Are Interviewer Effects on Interview Speed Related to Interviewer Effects on Straight-Lining Tendency in the European Social Survey? An Interviewer-Related Analysis

Are Interviewer Effects on Interview Speed Related to Interviewer Effects on Straight-Lining... Abstract It can be expected that interview speed and straight-lining tendency are related: a higher speed can lead to an increase in cognitive difficulty, which might be coped with by straight-lining; whereas, straight-lining can decrease response latency and hence increase the interview speed. Previous research shows that interview speed is influenced by the interviewer. Straight-lining tendency and other satisficing symptoms have also been shown to be subject to interviewer effects, albeit to a smaller extent. In the first step in this paper, we analyze interviewer effects on interview speed and straight-lining in one module of the European Social Survey, round 7. We then study the correlation between interview speed and straight-lining, without determining causality, and separate this into an interviewer-level and respondent-level correlation. The results confirm previous findings: a large intra-interviewer correlation coefficient for interview speed and a smaller intra-interviewer correlation coefficient for straight-lining. Further, the results show a significant positive correlation in nine out of fifteen countries at the respondent level and in five (different) countries at the interviewer level. The latter indicates that “fast” interviewers are the ones carrying out interviews during which more straight-lining occurs. 1. INTRODUCTION Research into the relationship between response time and straight-lining–two indicators of satisficing (Krosnick 1991; Krosnick, Narayan, and Smith 1996)–has mainly been carried out for online surveys, possibly because response times are easily collected for that mode (Heerwegh 2003) and because straight-lining is more likely to occur in self-administered surveys than in telephone or face-to-face equivalents (e.g., Fricker, Galesic, Tourangeau, and Yan 2005). As a consequence, existing research on this relationship is also limited to the effects of respondent and/or item characteristics. Studies show that speeding (high speed, short response time) is related to straight-lining (Conrad, Tourangeau, Couper, and Zhang 2011; Zhang and Conrad 2014; Revilla and Ochoa 2015). However, the relationship between response time and data quality in general is ambiguous. On the one hand, research generally shows that speeding is related not only to straight-lining but also to primacy effects (Malhotra 2008), less-detailed answers (Wells, Rao, Link, and Pierce 2012), and incoherence in answers (Revilla and Ochoa 2015). On the other hand, long response times have been linked to ambivalent attitudes (Johnson 2004), lack of knowledge (Heerwegh 2003), usability problems (McClamroch 2011), and, by contrast, engaged respondents (Crawford, Couper, and Lamias 2001). Response times, often referred to as response latencies, are a measurement of the cognitive effort made by the respondent to understand and answer a question (Bassili and Scott 1996). Hence, short response times can be a sign of satisficing or of a cognitive shortcut taken by the respondent (Krosnick 1991). However, short response times can also reflect a respondent’s accessible attitude (Grant et al. 2010), knowledge (Heerwegh 2003), or higher cognitive abilities (Bassili 1996; Yan and Tourangeau 2008; Couper and Kreuter 2013). However, Olson and Bilgen (2011) found longer interview durations for higher-educated respondents. This could be the consequence of more engagement, or of less satisficing, when completing a survey. Hence, respondent characteristics—in particular, cognitive abilities and motivation—together with item characteristics can influence response times in different directions. The relationship between speeding and data quality has also been found to be dependent on the respondent’s education level (Malhotra 2008; Zhang and Conrad 2014). In face-to-face surveys, the speed of an interview is jointly affected by the interviewer and the respondent. This makes the relationship between speed and data quality even more complex. Interviewers are responsible for the reading speed of the questions, the time given to respondents to answer, and the length of breaks. Respondents influence the interview speed through response latency. Based on the principle of standardized interviewing (Fowler and Mangione 1990), there is no reason why between-interviewer variation in interview speed should occur after controlling for proxies of a respondent’s cognitive capacity, such as age and education. In other words, after acknowledging that interviewers should adapt their speed to the respondents’ cognitive capacity, we should not expect differences between interviewers. However, research shows that in face-to-face surveys, interviewers do play a role in the interview speed (Hox 1994; Japec 2005; Olson and Peytchev 2007; Loosveldt and Beullens 2013a, 2013b). Cannell, Miller, and Oksenberg (1981, pp. 414–415) observed that the interviewer’s major goal seems to be to finish the interview as quickly as possible. This observation suggests that the interviewer can deliberately increase the interview speed. This tendency can vary between interviewers, depending upon, for instance, their motivation and workload. The tendency of interviewers to increase the interview speed could be an indication of their burden: interviewers having too many interviews to conduct or a lack of motivation to do their job properly (West and Blom 2016, p. 19). Olson and Peytchev (2007) found that interviewers with more experience perform their interviews faster and that the speed also increases over the fieldwork period. One can hypothesize that the questionnaire becomes familiar to the interviewer, or that the training in standardized interviewing focuses more on new interviewers than on the more experienced ones. Using data from the European Social Survey, Loosveldt and Beullens (2013b) confirmed these results and additionally found indications that the interview speed is related to the type and the complexity of the items. In summary, all the existing results of analyses of interview speed support the idea that interviewers can influence how fast the interview proceeds as much as respondents can. Literature about interviewer effects on straight-lining is sparser. Loosveldt and Beullens (2017) found that variation in the straight-lining behavior of respondents can be partly explained by interviewer effects, even after controlling for respondent characteristics (age, gender, and education level), the rank of the interview for the same interviewer (whether it was the first, second, third, etc., carried out by the interviewer during the data collection period), density (self-reported indication of population density), region, and the interviewer assessment of how well the respondent performed during the interview. Their results also show that in general—similarly to the interview speed—straight-lining increases with the rank of the interview. Further, interviewer effects have been shown to have an influence on other indicators of satisficing. Research shows that interviewers have an effect on acquiescence (Hox, De Leeuw, and Kreft 1991; Olson and Bilgen 2011). Olson and Bilgen (2011) found more acquiescence behavior among respondents who completed an interview with an experienced interviewer, but no relationship with interview speed. Further, they did not find any link between acquiescence and the gender, age, or education level of interviewers. They concluded that interviewer effects on acquiescence are caused by interviewer behavior: one can assume that more experienced interviewers establish a good relationship with respondents, which can lead to more acquiescence. Pickery and Loosveldt (2004) also found significant interviewer effects on the number of “don’t know” and “no opinion” answers. Lastly, Turner’s findings (2013; Turner, Sturgis, and Martin 2015) suggest that interviewers with a less accepting attitude to respondent refusal elicit more satisficing behavior among their respondents, whereas no links with interviewers’ demographic characteristics, skills, or personalities were found. Based on all these findings, a further analysis of straight-lining, as one indicator of satisficing, and its relationship to interview speed in face-to-face surveys seems relevant and appropriate. To the best of our knowledge, no previous research has been performed on how interviewers can affect the relationship between interview speed and straight-lining tendency in face-to-face surveys. This is the main research objective of the current paper. 2. RESEARCH OBJECTIVES In the first step, we separately study interviewer effects on interview speed (RQ1) and straight-lining (RQ2) in round 7 of the European Social Survey (ESS7). Although we expect to replicate previous findings regarding respondent and interviewer effects on interview speed, we consider this a necessary first step in order to validate our data and to underpin the importance of the further steps undertaken. It is also a relevant additional test of the external validity of previous results (Olson and Peytchev 2007; Loosveldt and Beullens 2013b). Next, we determine whether there is a relationship between interview speed and straight-lining tendency in face-to-face surveys (RQ3a). We do not hypothesize any causality in this regard, but expect these aspects of interviewer and response behavior to be positively correlated. Indeed, these two interview characteristics can be regarded as indications of interviewer and/or respondent satisficing behavior. Moreover, if a high interview speed is largely defined by the interviewer, this may increase the difficulty of answering the questions, and hence, trigger more satisficing in the respondent’s behavior, leading to straight-lining tendency, among other things. In reverse, straight-lining—which can be elicited by interviewer behavior—might increase the interview speed, with respondents not going through the full cognitive response process and answering faster. Accordingly, we consider that the interviewer behavior and the respondent behavior might both be responsible for the expected positive correlation between speed and straight-lining. Therefore, we also analyze how the correlation between interview speed and straight-lining tendency can be separated into the respondent level and the interviewer level (RQ3b). If a part of the correlation is attributed to the interviewer level, this would suggest that interviewer satisficing occurs, with some interviewers leading the interviews at a high speed and/or not motivating the respondents to give the best possible answers. For the three research objectives presented above, we also discuss possible differences between the countries participating in the ESS7. Overall, we do not expect differences between countries when they properly implement the survey according to the specifications of standardized interviewing. The comparison of different countries can, however, be considered an additional test of external validity. More information about the ESS7 is provided in the next section. 3. DATA: THE SEVENTH ROUND OF THE EUROPEAN SOCIAL SURVEY To answer our research questions, we use data from the European Social Survey round 7 (ESS 2014).1 Fifteen countries participated in the ESS7 using computer-assisted personal interviews (CAPI) (Austria, Belgium, Denmark, Estonia, Finland, France, Germany, Hungary, Ireland, the Netherlands, Norway, Slovenia, Sweden, Switzerland, and the United Kingdom) and six using (at least partly) paper-assisted interviews (the Czech Republic, Israel, Lithuania, Poland, Portugal, and Spain). We exclude the latter group from our analyses, as the recording of start and end times suffers from notable errors in this type of interview. In all the countries, samples were randomly selected—although sampling designs can vary (ESS 2015)—and should reach an effective sample size of 1,500 individuals. In addition to its goal of measuring changes in attitudes and beliefs in Europe, the ESS aims to reach the highest methodological standards to ensure valid cross-country comparison. Therefore, among many methodological projects conducted by the ESS, all participating countries using CAPI have had timers in their software since the ESS round 5. The timers measure the start (just before the first item, A1) and the end (immediately after the last item, F61) of the interview and also the start and end of each module (thematic group of items) in the questionnaire. This is the most detailed measurement of duration/speed available in the ESS round 7; no time data is available at the item level. The ESS questionnaire contains four core modules, largely similar from round to round, and two rotating modules. In round 7, immigration and health were the topics of the rotating modules. In the current analysis, we concentrate on the mean speed over Module B, the second module of the ESS core questionnaire, which covers political involvement, including political interest, trust, electoral, and other forms of participation; party allegiance; and socio-political orientation. We chose this module because we are interested in straight-lining in blocks or grids of more than three items. There are four such blocks in the whole round 7 questionnaire, three of which are in Module B, and the items in these blocks represent more than one-third of those in this module. Restricting the analysis to one module ensures that the time measurement is more closely related to the answering process for the items used to measure straight-lining. 4. MEASURING AND MODELING INTERVIEW SPEED: INTERVIEWER AND RESPONDENT EFFECTS The interview speed is defined as the number of items answered per minute. To calculate the mean interview speed over Module B, we use a variable that indicates the duration of the module in minutes (BINWTM). The recorded module interview length differs between countries, but ranges from 0 to 311 minutes. We remove extreme values from the analysis, defined as less than 5 minutes and more than 40 minutes. Loosveldt and Beullens (2013b) showed that language has a low impact on the interview speed and, hence, that the extreme values can be defined across countries. Moreover, we believe that the defined extreme values are unlikely, independent of the language, and must be due to technical or procedural issues. Lastly, the percentage of missing values before excluding extreme values varies between countries, from 0.8% (Finland) to 10.2% (the Netherlands). In the second step, we count the number of items in Module B that were presented to each respondent. The ESS questionnaire contains filter items, and as a consequence, the number of items asked can vary from one respondent to another. The presence of filters in Module B is limited and the difference between the minimum and maximum number of items is small (a minimum of 40 and a maximum of 44). Lastly, we calculate the mean module speed as the number of items asked in Module B, q, divided by the Module B length t in minutes for each respondent i: vi=qiti. Table 1 shows the mean Module B speed and the standard deviation over all respondents and for each country, after having removed the extreme values. The mean speed ranges from 3.22 items per minutes in Germany to 5.25 in Ireland. The mean speed we obtain is slightly lower than in the results obtained by Loosveldt and Beullens (2013b, p. 1425): three to four items per minutes compared with four to five items per minute. This could be due to different treatment of extreme values. Table 1. Descriptive Statistics of the Interview Speed in Module B Per Country Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 Table 1. Descriptive Statistics of the Interview Speed in Module B Per Country Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 To model the module speed and to evaluate the interviewer effects, we use a multilevel model in which respondents are clustered in interviewers. For each country separately, we assume that the module speed is not only determined by the respondents—as it would be for self-administered questionnaires—but also by the interviewers and that part of the variability can be explained by the differences between the interviewers at the second level. Because respondents are not randomly assigned to interviewers, but instead to interviewers who work in their vicinity, interviewer effects (intra-interviewer correlation) can be confounded with area effects (e.g., O’Muircheartaigh and Campanelli 1998; Schnell and Kreuter 2005; Vassallo, Durrant, and Smith 2017). The intertwining of interviewer and area effects is a complex problem, which can be partly circumvented by controlling for some relevant respondent characteristics. As already mentioned, existing research shows that a respondent’s age and level of education, which can serve as a proxy for cognitive abilities, influence the interview speed, although results are not consistent with regard to the direction of this relationship (e.g., Olson and Bilgen 2011; Loosveldt and Beullens 2013a, 2013b). It has also been shown that the rank of the interview plays a role in the interview speed (Olson and Peytchev 2007; Loosveldt and Beullens 2013a). Hence, we control for these characteristics, as well as for these characteristics: respondent’s gender; whether or not the interview was conducted in the respondent’s mother tongue2 (Lang); the population density as described by the respondent on a five-point scale (Dens); the region based on counties, provinces, or other subnational entities (Reg); the interviewer assessment on a five-point scale of whether the respondent answered to the best of their abilities (Abil); and of whether the respondent understood the questions (Und). These are characteristics that we believe may influence the interview speed. The following model (Model A) is a random-intercept, fixed-slopes design in which we control for respondent and interview characteristics: vi,j=β0,j+β1Womani,j+β2Langi,j+β3Educ1i,j+β4Educ2i,j+β5Agei,j+β6Ranki,j+β7Densi,j+β8Regi,j+β9Abili,j+β10Undi,j+ɛi,j, with β0,j=γ0,0+u0,jwhere β0,j is the random intercept, βk from k = 1 to k = 6, are the fixed effects, γ0,0 ⁠, the overall intercept; ɛi,j∼N0,σw2; and u0,j∼N(0,σb2) ⁠. The intra-interviewer correlation coefficient ρv that represents the portion of variance in speed that is explained by the variation between interviewers is given by: ρv=σb2σw2+σb2. For each country for Module B in the ESS7, table 2 displays the overall intercept γ0,0 ⁠, the fixed slope for the dependent variables gender β1 ⁠, whether or not the interview was taken in the respondent’s mother tongue (Lang) β2 ⁠, level of education (upper-secondary β3 and tertiary β4 ⁠), age β5 ⁠, rank β6 ⁠, population density (Dens) β7 ⁠, whether the respondent answered to the best of their ability (Abil) β9 ⁠, whether the respondent understood the questions (Und) β10 ⁠, the between variance σb2 and the residual variance σw2 ⁠, and the intra-class correlation coefficient ρv ⁠. There are a number of categorical variables: “Woman” (0 = man, 1 = woman); “Lang” (0 = language of interview is not mother tongue, 1 = language of interview is mother tongue); two dummies for education based on ISCED level, “Educ1” (1 = ISCED level 3 or 4, 0 = other); and “Educ2” (1 = ISCED level 5, tertiary, 0 = other). Age and rank are continuous variables. The population density, the respondent abilities, and the respondent’s understanding are considered continuous. We do not display the effects of the region; these are lengthy to report and often not significant at the 0.05 level, except for a few regions in Belgium, France, Germany, Hungary, Ireland, and Norway. All the codes for data preparation and modelling can be found in the supplementary materials. Table 2. Parameter Estimates of Model A for Module Speed (ESS7) Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 * p < 0.10, **p < 0.05. Table 2. Parameter Estimates of Model A for Module Speed (ESS7) Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 * p < 0.10, **p < 0.05. The sample sizes vary from 1,158 (after removing missing durations and extreme values) in Slovenia, with the lowest number of interviewers (60), to 2,945 in Germany, with the highest number of interviewers (266). The average number of respondents per interviewer (i.e., the workload) varies from 9.9 in the United Kingdom to 23.4 in Switzerland. The overall intercept also varies substantially between countries, ranging from 2.42 items per minute in Sweden to 6.14 items per minute in Slovenia. In regard to RQ1, the intra-interviewer correlation coefficients are all above 8% (Finland), with a maximum value of 37% (Switzerland). Therefore, interviewer effects on speed cannot be neglected, regardless of the country. Indeed, Groves (1989, p. 364) notes: “Most other surveys have an average ρint between 0.01 and 0.02.” Although acceptable values for the intra-interviewer coefficient depend on the number of interviewers, the sample size, and interaction with area effects, we consider interviewer effects as substantial whenever the intra-interviewer correlation coefficient exceeds 0.05. For comparison, in round 7 the intra-interviewer correlation coefficient for the respondent age varies between 0.00 in Belgium and 0.06 in Estonia. These findings confirm our expectations based on previous research (Hox 1994; Japec 2005; Olson and Peytchev 2007; Loosveldt and Beullens 2013b). Moreover, the results underline the importance of analyzing the link between the effect of interviewers on interview speed and indicators of data quality such as straight-lining (RQ3b). Although not the core interest in this paper, we briefly discuss the effects of the independent variables in Model A. Women have slower interviews in Finland, Germany, Norway, Sweden, Switzerland, and the United Kingdom. When the interview language is the respondent’s mother tongue, the interview speed is mostly higher, as would be expected (the exceptions are Austria and Slovenia, but non-significant), although the effect is not consistently significant. This might, however, depend on the size of the group for which the mother tongue is not the interview language. The direction of the effect of education is not clear, and the effect itself is quite often not significant. This could be the result of a mixed effect of higher-educated people having greater cognitive abilities, but also using more cognitive effort to give an answer. Having an upper-secondary (Educ1) and/or tertiary (Educ2) level of education significantly increases the speed, compared with lower-secondary education or below in two countries (Finland and Switzerland) and significant decreases in speed in Ireland. However, the pattern is not necessarily linear; compared with people having a lower education level, the increase in speed for people with a tertiary level of education does not always exceed that for people with an upper-secondary level of education. The interview speed also decreases with the respondent’s age in all countries (Austria not significant) and increases with the rank of the interview (Austria, Finland, and Germany not significant). This supports the idea that there is a “learning” effect for interviewers, which is in line with previous findings (Olson and Peytchev 2007; Loosveldt and Beullens 2013b). The population density almost never has a significant effect on the interview speed, except for Norway where people in less dense areas have slower interviews. The interview speed decreases when the respondent more frequently answers to the best of their ability, a proxy for motivation (only significant in Belgium, Estonia, and Finland), whilst the speed increases when the respondent more frequently understands the questions. However, these two variables are reported by the interviewers and could therefore also suffer from interviewer effects. The model (Model A) was also run without the independent variables at the respondent level. The resulting proportions of variance explained by the variance between interviewers, without controlling for respondent characteristics, are close to those shown for Model A. 5. MEASURING AND MODELING STRAIGHT-LINING TENDENCY: INTERVIEWER AND RESPONDENT EFFECTS Although the definition of straight-lining is clear—the tendency to give the same answer to items regardless of the content of the item—relevant literature suggests different possible ways to measure this (Mulligan, Krosnick, Smith, Green, and Bizer 2001; Chang and Krosnick 2009; Zhang and Conrad 2014; Loosveldt and Beullens 2017). The first point for discussion concerns the use of a homogeneous or a heterogeneous set of items. A heterogeneous set of items may be seen as better suited because it might reasonably be expected that the respondent’s answer will differ from one item to another if the items have a low inter-correlation. Giving the same answer to a homogenous set of items that are nuances of the same topic may reflect true values, rather than being caused by straight-lining. However, in the current analysis we are limited to homogeneous sets of items, because the ESS questionnaire does not contain a suitable set of heterogeneous items. The second discussion point concerns pure straight-lining versus straight-lining tendency. Zhang and Conrad (2014) studied pure patterns in an online survey, where exactly the same answer was given for all the items in a grid. We argue, however, that this is more likely to occur in an online questionnaire than in a face-to-face interview and more likely to occur for shorter sets of items than for longer sets. The choice to only consider pure straight-lining does not shed any light on straight-lining tendencies, which are also important and which we expect to find in face-to-face surveys with longer sets of items. We therefore measure the straight-lining tendency by examining the percentage of answers that are exactly the same as the previous one in a set, including refusals, “don’t know,” and “no answer” (Loosveldt and Beullens 2017). For example, examining the following answers to a set of two times four items which are considered as a whole: 3, 3, 2, 1 and 1, 1, 4, 2, the percentage of straight-lined answers is 33.3% (two out of six, with one identical subsequent answer in each set, ignoring the transition between sets). There is a potential maximum of three straight-lined answers in each set, as the first item obviously has no previous answer to compare it with. The measurement can, in this example, only take seven values (0%, 16.7%, 33.3%, 50%, 66.6%, 83.3%, or 100%). Hence 33.3% may seem large, although looking at the answers given in the example, straight-lining is not obvious. The measurement is better suited and more robust for more blocks and more items in blocks (as in the ESS7 Module B). Other ways to measure straight-lining could, however, be considered. One possibility would be to examine the standard deviation of the scores for each block and each respondent. A low standard deviation would be a sign of low discrimination or non-discrimination in the answers in one block. However, a switching pattern—for instance, 2, 3, 2, 3 and 4, 5, 4, 5—would also result in a low standard deviation, although these answers do not follow a straight-lining pattern. Another possibility would be to use Mulligan’s score (Mulligan et al. 2001; Chang and Krosnick 2009). Mulligan’s score is a distance metric, measuring the average square root of the absolute difference between any two answers by the same respondent in a block of items, or n2-1∑i=1n∑i'>inxi-xi' ⁠, where n is the number of items in the grid and x is the answer of the respondent to the item. The standard deviation method and Mulligan’s score are, however, distance measurements that evaluate non-discrimination or low discrimination and do not reflect the sequence pattern. Moreover, based on the sensitivity analysis detailed by Loosveldt and Beullens (2017), we believe that the results are in practice independent of the chosen measurement. Table 3 presents the three blocks or sets of items and their characteristics that we use to measure straight-lining tendency. We chose blocks that contain more than three items, and these all contain 11-point scale items, which does not allow us to study scale effects. The total number of items in these blocks is seventeen, which represents more than one-third of all the items in Module B. The maximum number of straight-lined answers is fourteen (pure straight-lining). This means that the measurement of the tendency for straight-lining can take fifteen values (0%, 100*1/14%, 100*2/14% … 100*13/14%, 100%). Table 3. Blocks of Items Used in the Calculation of the Percentage of Straight-Lined Answers Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Table 3. Blocks of Items Used in the Calculation of the Percentage of Straight-Lined Answers Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Table 4 displays, for each country, the mean straight-lining tendency measurement and the standard deviation over all respondents. The mean straight-lining tendency ranges from 19.8% (slightly less than three straight-lined items) in Germany to 28.6% (between four and five straight-lined items) in Hungary. Table 4. Descriptive Statistics of the Straight-Lining Tendency Measurement in Module B Per Country Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 Table 4. Descriptive Statistics of the Straight-Lining Tendency Measurement in Module B Per Country Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 In a similar way to the interview speed, interviewer effects on straight-lining can be modeled by a multilevel model to separate respondent variance from interviewer variance. We control for the same respondent characteristics as the ones we use in relation to the interview speed: age and level of education (as proxies for cognitive abilities), the rank of the interview, gender, whether or not the interview was in the respondent’s mother tongue, population density, region, whether the respondent answered to the best of their ability, and whether the respondent understood the questions. The following model (Model B) is a random-intercept, fixed-slopes model in which we control for respondent and interview characteristics: stri,j=ϕ0,j+ϕ1Womani,j+ϕ2Langi,j+ϕ3Educ1i,j+ϕ4Educ2i,j+ϕ5Agei,j+ϕ6Ranki,j+ϕ7Densi,j+ϕ8Regioni,j+ϕ9Abili,j+ϕ10Undi,j+ωi,j, with ϕ0,j=μ0,0+v0,j ⁠.where ϕ0,j is the random intercept, ϕk from k = 1 to k = 6 are the fixed effects, μ0,0the overall intercept, ωi,j∼N(0,τw2) ⁠, and v0,j∼N(0,τb2) ⁠. The intra-interviewer correlation coefficient ρstr ⁠, which represents the proportion of variance in straight-lining tendency that is explained by variation between interviewers is given by: ρstr=τb2τw2+τb2 Table 5 displays, for each country, the overall intercept μ0,0 ⁠; the fixed slope for the dependent variables; gender ϕ1 ⁠; whether the interview was in the respondent’s mother tongue ϕ2 ⁠; level of education ϕ3; ϕ4 ⁠; age ϕ5 ⁠; rank ϕ6 ⁠; population density ϕ7 ⁠; whether the respondent answered to the best of their ability ϕ9 ⁠; whether the respondent understood the questions ϕ10 ⁠; the between variance τb2 and the residual variance τw2 ⁠; and the intra-class correlation coefficient ρstr for Module B in the ESS7. Again, to stay concise the region effect is not displayed, but is significant for a few regions in Belgium, France, Hungary, Germany, Ireland, Slovenia, and Sweden. Table 5. Parameter Estimates of Model B2 for Straight-Lining Tendency (ESS7) Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 * p < 0.10, **p < 0.05. Table 5. Parameter Estimates of Model B2 for Straight-Lining Tendency (ESS7) Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 * p < 0.10, **p < 0.05. The overall intercept varies from 16.25% of straight-lined answers in Germany (between two and three items out of the fourteen) to 39.20% in Hungary (close to six items out of the fourteen). In countries in which gender has a significant effect at the 0.05 level (Austria, Belgium, Finland, Germany, Hungary, Ireland, Slovenia, Sweden, and the United Kingdom), women tend to straight-line more often. Taking the interview in the mother tongue reduces the percentage of straight-lined answers in Austria, Germany, Norway, Slovenia, and Sweden. Generally, the effect of (higher) education is negative (a lower straight-lining tendency). Further, the percentage of straight-lined answers increases with age (with the exception of the United Kingdom), and the rank of the interview has no effect on the tendency to straight-line (with the exception of Belgium). The population density also does not have an effect on straight-lining. Both higher respondent motivation (Abil) and understanding (Und) lead to fewer straight-lined answers. The intra-class correlations (ICCs) are lower than 0.05 in Denmark, Finland, France, Germany, the Netherlands, Norway, Slovenia, Sweden, Switzerland, and the United Kingdom, and lower than 0.10 in Belgium. In Austria, Estonia, Hungary, and Ireland, the ICCs are above 0.13. These differences between countries are in line with previous findings (Loosveldt and Beullens 2017). Moreover, the results suggest that the influence of the interviewer on the straight-lining behavior of the respondent is smaller than the influence of the interviewer on the interview speed. This is not surprising, since the interviewer impact on straight-lining is less direct (through motivation and task difficulty) than on interview speed, in which the interviewer plays an active role. Again, controlling for respondent characteristics has only a small effect on the proportion of variance explained at the interviewer level, suggesting that the interviewer effects are larger than the clustering effects due to homogenous groups of respondents in particular areas. 6. SEPARATING THE CORRELATION BETWEEN INTERVIEW SPEED AND STRAIGHT-LINING TENDENCY INTO THE INTERVIEWER LEVEL AND THE RESPONDENT LEVEL Our main research questions (RQ3a and RQ3b) concern the correlation between straight-lining tendency and interview speed, and the effect of interviewers on this correlation. Accordingly, we are interested not only in separating interview speed and straight-lining tendency variances into the respondent level and the interviewer level, but also in the effect of the respondents and the interviewers on the relationship between these two proxies of satisficing. The correlations within countries at the respondent level—without taking into account the nesting within interviewers (RQ3a)—between interview speed and straight-lining tendency are presented in table 6 (column correlations). Eight of the countries display a small but significant correlation (not exceeding 0.13) between straight-lining tendency and interview speed, suggesting that such a relationship may exist. Moreover, most correlations are—as expected—positive, indicating that a faster interview speed is related to an increased straight-lining tendency. We can hypothesize different reasons for this low correlation, as the interaction between the interviewer and the respondent is complex. First, the speed is not measured specifically over the items in blocks, but over the whole module, which may reduce the amplitude of the correlation. Moreover, the interviewer and respondent both affect the speed and the straight-lining behavior, but the interviewer effects on speed are larger and more straightforward (a direct role through reading speed). Lastly, the relationship of speed to data quality is ambiguous. On the one hand, a high speed can be a sign of satisficing and can lead, for instance, to straight-lining; but it can also be a sign of the respondent’s knowledge and understanding. On the other hand, a lower speed can be a sign of (too high) cognitive difficulty, but also of more cognitive effort to provide an answer. Finland has a negative significant correction between interview speed and straight-lining. This is an unexpected result. Although we do not have recordings that would be necessary to understand this phenomenon, we can hypothesize that straight-lining is a satisficing behavior due to the cognitive difficulty the respondent experiences and that the interviewer slows down the interview to help the respondent. Table 6. Correlations between Straight-Lining Tendency and Interview Speed and the Separation into Within and Between Components Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* p < 0.10 and *p < 0.05 Table 6. Correlations between Straight-Lining Tendency and Interview Speed and the Separation into Within and Between Components Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* p < 0.10 and *p < 0.05 In the next step we specify a multivariate multilevel model (Heck and Thomas 2015, Chapter 6). With this model, we can separate the relationship between interview speed and straight-lining tendency into the interviewer and the respondent level without specifying the direction of this relationship. Therefore, we specify both interview speed and straight-lining tendency as dependent variables. To estimate the variance and covariance between interview speed and straight-lining tendency at both levels, we consider the following multivariate model, controlling for respondent characteristics (Models A and B) as illustrated in figure 1. Figure 1. View largeDownload slide Multivariate multilevel model with explanatory variables at the respondent level. Figure 1. View largeDownload slide Multivariate multilevel model with explanatory variables at the respondent level. We then estimate the following model: v(i,j)stri,j=β0,jϕ0,j+00β00ϕv(i,j)str(i,j)Xi,j+ɛi,jωi,j β0,jϕ0,j=γ0,0μ0,0+u0,jv0,j with covariance matrices ΨW=σw2ϑv,strϑv,strτw2 at the within level and ΨB=σb2αv,strαv,strτb2 at the between level. Conforming to models A and B, the random intercepts for the interview speed and straight-lining tendency are given by respectively β0,j and ϕ0,j ⁠, and the overall intercepts are given by γ0,0 and μ0,0 ⁠. The regression coefficient vectors for interview speed and straight-lining tendency are denoted by β=(β1β2β3β4β5β6β7β8β9β10) and ϕ=(ϕ1ϕ2ϕ3ϕ4ϕ5ϕ6ϕ7ϕ8ϕ9ϕ10) ⁠. The symbol Xi,j denotes the vector of respondent characteristics for which we control. The respondent-level variances are given by σw2 and τw2, and the interviewer-level variances are given by σb2 and τb2 ⁠. The main parameters we are interested in are the covariances between interview speed and straight-lining tendency at the respondent level (within) ϑv,str and at the interviewer level (between) αv,str ⁠. Correlations are calculated using standardized covariances. The resulting correlations between straight-lining tendency and interview speed separated into the interviewer and the respondent level, after controlling for respondent characteristics, are displayed in the two last columns of table 6 (Corr. within and Corr. between). The results of the separation into the respondent level (within) and the interviewer level (between) of the correlation, addressing our last research question (RQ3b), are interesting. In nine countries (Belgium, Denmark, France, Ireland, the Netherlands, Slovenia, Sweden, Switzerland, and the United Kingdom), a small but significant positive correlation is found at the respondent level between straight-lining tendency and speed. This indicates a positive relationship between speed and straight-lining independent of the interviewer, probably meaning that these are both indicators of respondent satisficing. The correlation between speed and straight-lining remains significant at the respondent level in six countries (Belgium, Ireland, the Netherlands, Slovenia, Switzerland, and the United Kingdom) out of the eight in which the correlation is significant at the interview level (without decomposition). This may be an indication that the relationship between interview speed and straight-lining is influenced by the respondent, the interviewer, and their interaction. Moreover, the correlation is positive and significant at the interviewer level in five countries (Estonia, Ireland, Slovenia, Switzerland, and the United Kingdom). In these countries, the interviewer behavior conjointly influences the interview speed and the straight-lining behavior. Given that speed and straight-lining are both indicators of satisficing, we may hypothesize that the interviewer behavior induces satisficing behavior from the respondent. Although we can observe that the relationship between straight-lining tendency and interview speed is not very strong overall, the results in table 6 provide empirical proof that this relationship does exist at the interviewer level in some countries. 7. DISCUSSION The aim of this paper was to validate previous results, which show large interviewer effects on interview speed and smaller but substantial interviewer effects on straight-lining tendency, at least in some countries. The analysis was then taken a step further by examining the correlation between interview speed and straight-lining tendency, which are both indicators of satisficing. This correlation was then separated into the interviewer level and the respondent level using a multilevel multivariate model. The analyses were performed using data from fifteen countries that participated in the ESS7 using CAPI. Our results confirm the findings in previous research (Olson and Peytchev 2007; Loosveldt and Beullens 2013a, 2013b, 2017), offering support for external validation. Interviewer effects on interview speed in the ESS7 are large, ranging from 0.08 to 0.37 across countries, even after controlling for respondent characteristics. It is notable that the variability in interview speed is explained to a major extent by differences between interviewers. Interview speed seems to be a powerful variable to detect deviations from standardized interviews and to analyze other characteristics of data quality. The interviewer effects on straight-lining tendency are considerably smaller, ranging from 0.00 to 0.21, but still substantial in some countries, showing again the possible impact of interviewers on a data quality indicator for which the respondent and response behavior are usually considered responsible. Turning to the relationship between interview speed and straight-lining tendency, the correlation between the two indicators of satisficing is low, but positive and significant in seven of the fifteen countries examined. This supports the hypothesis that high interview speed and straight-lining are related to each other, probably through both interviewer and respondent satisficing behavior. Separating the correlation into interviewer and respondent levels within countries, however, shows a significant correlation at the respondent level in nine countries. In these, fast respondents—who are possibly giving straight-lined answers—might influence the interviewer to increase the interview speed, because the respondent wants to finish the interview quickly. More importantly, the interviewer-level correlation is significant in five countries. This result implies that in these countries, interviewers who carry out interviews at a high speed are also the ones who are responsible for interviews with a higher straight-lining tendency and vice versa. Slovenia, Switzerland, and the United Kingdom belong to the countries with a significant correlation at the interviewer level. This is surprising given that the intra-interviewer correlation of the percentage straight-lined answers is lower than 0.05 in these countries. This means that the small differences between interviewers in straight-lining tendency among their respondents is strongly linked to the speed at which they conducted the interview. It would be therefore interesting to look at other quality measures, such as acquiescence or the use of “don’t know” answers, and their relationship with interview speed at the respondent and interviewer level to check whether the same results can be observed. This study has some limitations. First, given the non-experimental nature of the data, we are not able to draw any conclusions about the causality of the relationship between interview speed and straight-lining tendency. An experiment in which interviewers are randomly assigned to two groups could address this issue. One group (half of the interviewers chosen randomly) would be instructed and trained to read the questions at a higher speed and to leave only short breaks, whilst the other group would be instructed and trained to read the questions more slowly and to leave longer breaks. If more straight-lined answers were found among the respondents interviewed by the first group than amongst the respondents interviewed by the second, interview speed could be shown to cause straight-lining. The questionnaire would, however, need to contain many blocks of items. Further, straight-lining only reflects one aspect of data quality. Different indicators could be considered, such as consistency or item nonresponse. One possibility would be to build a satisficing indicator such as the one constructed by Kaminska, McCutcheon, and Billiet (2010), and to study its relationship to interview speed at the respondent and the interviewer level. Lastly, a measurement of interview speed that is more directly linked to the measurement of straight-lining tendency would be preferable. Timers measuring the start and the end of blocks of items would probably provide a stronger proof of the relation between speed and straight-lining. To conclude, our analysis offers empirical proof of deviations from standardized interviewing and of the effect that interviewers may have on the quality of data. This brings into question the interviewer’s role in face-to-face surveys. Interviewers should facilitate the completion of the questionnaire and motivate respondents to optimize the response process and consequently give the best answers. However, interviewers rushing through the process can increase the difficulty of the task and reduce the data quality. Interview speed as a characteristic of the interviewer seems to be related to data quality indicators such as straight-lining and could, therefore, be used for fieldwork monitoring. Moreover, it might be necessary to integrate interviewers into analyses concerning response styles. Given that previous research (Olson and Peytchev 2007) shows that more-experienced interviewers perform faster interviews, the results imply that retraining these interviewers in standardized interviewing techniques may be necessary. The results also prove the importance of interviewer motivation, not only to gain the respondents’ participation but also to perform high quality interviews at the right speed rather than rushing through. Hence, interviewer training should also focus on interview speed, on top of focusing on doorstep interaction and keeping a neutral attitude. Further research should investigate in greater detail the relationship between interview speed and interaction characteristics. What aspects of interviewer behavior are responsible for the large differences between interviewers in interview speed? Which interaction characteristics have the largest effect on interview speed and/or on straight-lining tendency? The behavior of the interviewer can influence both the interview speed and the straight-lining tendency. Interviewer behavior characteristics such as whether the question is read exactly as written, the number of probes (Ongena and Dijkstra 2006), whether the interviewer apologetically pursues the reading of the remaining response options when the respondent interrupts with an answer before the end, whether the interviewer repeats the question when an inappropriate answer is given (Garbarski, Schaeffer, and Dykema 2016), or whether the interviewer gives clarification to the question (Schober, Conrad, Dijkstra, and Ongena 2012) can influence the interaction with the respondents and hence the interview speed. The respondent’s engagement and motivation (Garbarski et al. 2016), but also understanding of the question or discomfort with the answer, can also influence the interview speed. The use of “um” or laughter, as well as the fluency of speech or gaze aversion, can be signs of difficulty answering a question (Schober et al. 2012). To perform the required analysis, interview recordings would be needed or timers that differentiate between when the interviewer or the respondent is speaking. However, the best coding schemes to choose for the recordings and their level of precision would need to be defined (Ongena and Dijkstra 2006). Lastly, we believe that the correlation at the interviewer level between interview speed and straight-lining tendency discovered in this paper highlights the importance of interview speed as a quality indicator that is relatively easy to monitor during the fieldwork, but very little used. It seems worthwhile to investigate the relationship between interview speed and other data quality indicators. Supplementary Materials Supplementary materials are available online at http://www.oxfordjournals.org/our_journals/jssam/. Footnotes 1 http://www.europeansocialsurvey.org/data/download.html?r=7, last accessed October 2017. 2 We do not have information on the mother tongue itself, but we use the primary language spoken at home as a proxy. REFERENCES Bassili J. ( 1996 ), “The ‘How’ and ‘Why’ of Response Latency Measurement in Survey Research,” in Answering Questions: Methodology for Determining Cognitive and Communicative Processes in Survey Research , eds. Schwartz N. , Sudman S. , pp. 319 – 346, San Francisco : Jossey-Bass Publishers . Bassili J. N. , Scott B. S. ( 1996 ), “ Response Latency as a Signal to Question Problems in Survey Research ,” Public Opinion Quarterly , 60 , 390 – 399 . Google Scholar Crossref Search ADS Cannell C. F. , Miller P. V. , Oksenberg L. ( 1981 ), “ Research on Interviewing Techniques ,” Sociological Methodology , 12 , 389 – 437 . Google Scholar Crossref Search ADS Chang L. , Krosnick J. ( 2009 ), “ National Surveys via RDD Telephone Interviewing versus the Internet. Comparing Sample Representativeness and Response Quality ,” Public Opinion Quarterly , 73 , 641 – 678 . Google Scholar Crossref Search ADS Conrad F. G. , Tourangeau R. , Couper M. P. , Zhang C. ( 2011 ), “Interactive Interventions in Web Surveys Can Increase Response Accuracy,” paper presented at the 2011 Annual Meeting of the American Association for Public Opinion Research, Phoenix, AZ. Couper M. P. , Kreuter F. ( 2013 ). “ Using Paradata to Explore Item Level Response Times in Surveys ,” Journal of the Royal Statistical Society: Series A , 176 , 271 – 286 . Google Scholar Crossref Search ADS Crawford S. C. , Couper M. P. , Lamias M. J. ( 2001 ), “ Web Surveys ,” Social Science Computer Review , 19 , 146 – 162 . Google Scholar Crossref Search ADS ESS ( 2014 ), ESS Round 7: European Social Survey Round 7 Data (2014) Data file ( 1st ed. ), Norway : Norwegian Social Science Data Services, Data Archive and distributor of ESS data for ESS ERIC . ESS ( 2015 ), ESS Round 7: European Social Survey (2015) ESS-7 2014. Data Documentation Report (1st ed.) , Bergen : European Social Survey Data Archive, Norwegian Social Science Data Services for ESS ERIC Fowler F. J. , Mangione T. ( 1990 ), Standardized Survey Interviewing. Minimizing Interviewer‐Related Error , Newbury Park : SAGE Publications . Fricker S. , Galesic M. , Tourangeau R. , Yan T. ( 2005 ), “ An Experimental Comparisons of Web and Telephone Surveys ,” Public Opinion Quarterly , 69 , 370 – 392 . Google Scholar Crossref Search ADS Garbarski D. , Schaeffer N. C. , Dykema J. ( 2016 ), “ Interviewing Practices, Conversational Practices and Rapport: Responsiveness and Engagement in the Standardized Survey Interview ,” Sociological Methodology , 46 , 1 – 38 . Google Scholar Crossref Search ADS PubMed Grant J. T. , Mockabee S. T. , Monson J. Q. ( 2010 ), “ Campaign Effect on the Accessibility of Party Identification ,” Political Research Quarterly , 63 , 529 – 543 . Google Scholar Crossref Search ADS Groves R. M. ( 1989 ), Survey Error and Survey Costs , Hoboken, NJ : John Wiley & Sons . Heck R. H. , Thomas S. L. ( 2015 ), An Introduction to Multilevel Modeling Techniques: MLM and SEM Approaches using Mplus , ( 3rd ed.) , New York : Taylor and Francis . Heerwegh D. ( 2003 ), “ Explaining Response Latencies and Changing Answers Using Client-Side Paradata from a Web Survey ,” Social Science Computer Review , 21 , 360 – 373 . Google Scholar Crossref Search ADS Hox J. ( 1994 ), “ Hierarchical Regression Models for Interviewer and Respondent Effects ,” Sociological Methods and Research , 22 , 300 – 318 . Google Scholar Crossref Search ADS Hox J. , De Leeuw E. , Kreft I. ( 1991 ), “The Effect of Interviewer and Respondent Characteristics on the Quality of Survey Data: A Multilevel Model,” in Measurement Errors in Surveys , eds. Biemer P. P. , Groves R. M. , Lyberg L. E. , Mathiowetz N. A. , Sudman S. , pp. 439 – 461 , New York : John Wiley . Japec L. ( 2005 ), “Interviewer Burden and its Effects on Data Quality in the Swedish Part of the European Social Survey (ESS),” in Quality Issues in Interview Surveys, Some Contributions , ed. Japec L. , Stockholm : Department of Statistics Stockholm University . Johnson M. ( 2004 ), “ Timepieces: Components of Survey Question Response Latencies ,” Political Psychology , 25 , 679 – 702 . Google Scholar Crossref Search ADS Kaminska O. , McCutcheon A. L. , Billiet J. ( 2010 ), “ Satisficing among Reluctant Respondents in a Cross-National Context ,” Public Opinion Quarterly , 74 , 956 – 984 . Google Scholar Crossref Search ADS Krosnick J. A. ( 1991 ), “ Response Strategies for Coping with the Cognitive Demands of Attitude Measures in Surveys ,” Applied Cognitive Psychology , 5 , 213 – 236 . Google Scholar Crossref Search ADS Krosnick J. A. , Narayan S. , Smith W. R. ( 1996 ), “ Satisficing in Surveys: Initial Evidence ,” New Directions for Evaluation , 1996 , 29 – 44 . Google Scholar Crossref Search ADS Loosveldt G. , Beullens K. ( 2013a ), “ How Long Will It Take? An Analysis of Interview Length in the Fifth Round of the European Social Survey ,” Survey Research Methods , 7 , 69 – 78 . Loosveldt G. , Beullens K. ( 2013b ), “ The Impact of Respondents and Interviewers on Interview Speed in Face-to-Face Interviews ,” Social Science Research , 42 , 1422 – 1430 . Google Scholar Crossref Search ADS Loosveldt G. , Beullens K. ( 2017 ), “ Interviewer Effects on Non-Differentiation and Straightlining in the European Social Survey ,” Journal of Official Statistics, Special Issue on Total Survey Error (TSE) , 33 , 409 – 426 . Malhotra N. ( 2008 ), “ Completion Time and Response Order Effects in Web Surveys ,” Public Opinion Quarterly , 72 , 914 – 934 . Google Scholar Crossref Search ADS McClamroch K. J. ( 2011 ), “ Evaluating the Usability of Personal Digital Assistants to Collect Behavioral Data on Adolescent with Paradata ,” Field Methods , 23 , 219 – 242 . Google Scholar Crossref Search ADS Mulligan K. , Krosnick J. , Smith W. , Green M. , Bizer G. ( 2001 ), Non-Differentiation on Attitude Rating Scales: A Test of Survey Satisficing Theory, unpublished manuscript, Stanford University. Olson K. , Bilgen I. ( 2011 ), “ The Role of Interviewer Experience on Acquiescence ,” Public Opinion Quarterly , 75 , 99 – 114 . Google Scholar Crossref Search ADS Olson K. , Peytchev A. ( 2007 ), “ Effect of Interviewer Experience on Interview Speed and Interviewer Attitudes ,” Public Opinion Quarterly , 71 , 273 – 286 . Google Scholar Crossref Search ADS O’Muircheartaigh C. , Campanelli P. ( 1998 ), “ The Relative Impact of Interviewer Effects and Sample Design Effects on Survey Precision ,” Journal of the Royal Statistical Society: Series A , 1 , 63 – 77 . Google Scholar Crossref Search ADS Ongena Y. P. , Dijkstra W. ( 2006 ), “ Methods of Behavior Coding of Survey Interviews ,” Journal of Official Statistics , 22 , 419 – 451 . Pickery J. , Loosveldt G. ( 2004 ), “ A Simultaneous Analysis of Interviewer Effects on Various Data Quality Indicators with Identification of Exceptional Interviewers ,” Journal of Official Statistics , 20 , 77 – 89 . Revilla M. , Ochoa C. ( 2015 ), “ What Are the Links in a Web Survey among Response Time, Quality, and Auto-Evaluation of the Efforts Done ?” Social Science Computer Review , 33 , 97 – 114 . Google Scholar Crossref Search ADS Schnell R. , Kreuter F. ( 2005 ), “ Separating Interviewer and Sampling-Point Effects ,” Journal of Official Statistics , 21 , 389 – 410 . Schober M. F. , Conrad F. G. , Dijkstra W. , Ongena Y. P. ( 2012 ), “ Disfluencies and Gaze Aversion in Unreliable, Responses to Survey Questions ,” Journal of Official Statistics , 28 , 555 – 582 . Turner M. ( 2013 ), New Perspectives on Interviewer-Related Error in Surveys: Application of Survey Paradata , Southampton : University of Southampton, Faculty of Social and Human Sciences . Turner G. , Sturgis P. , Martin D. ( 2015 ), “ Can Response Latencies Be Used to Detect Survey Satisficing on Cognitively Demanding Questions ?” Journal of Survey Statistics and Methodology , 3 , 89 – 108 . Google Scholar Crossref Search ADS Vassallo R. , Durrant G. , Smith P. ( 2017 ), “ Separating Interviewer and Area Effects by using a Cross-Classified Multilevel Logistic Model: Simulation, Findings and Implications for Survey Designs ,” Journal of the Royal Statistical Society: Series A , 180 , 531 . Google Scholar Crossref Search ADS Wells T. , Rao K. , Link M. W. , Pierce C. ( 2012 ), “Flagging Speeders in a Multi-Mode (Mobile and Online) Survey,” paper presented at the 2012 Annual Meeting of the Midwest Chapter of the American Association for Public Opinion Research, Chicago, IL. West B. T. , Blom A. G. ( 2016 ), “ Explaining Interviewer Effects: A Research Synthesis ,” Journal of Survey Statistics and Methodology , 5, 175 – 211 . Yan T. , Tourangeau R. ( 2008 ), “ Fast Times and Easy Questions: the Effects of Age, Experience and Question Complexity on Web Survey Response Times ,” Applied Cognitive Psychology , 22 , 51 – 68 . Google Scholar Crossref Search ADS Zhang C. , Conrad F. G. ( 2014 ), “ Speeding in Web Surveys: The Tendency to Answer Very Fast and Its Association with Straightlining ,” Survey Research Methods , 8 , 127 – 135 . © The Author 2017. Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Survey Statistics and Methodology Oxford University Press

Are Interviewer Effects on Interview Speed Related to Interviewer Effects on Straight-Lining Tendency in the European Social Survey? An Interviewer-Related Analysis

Loading next page...
 
/lp/ou_press/are-interviewer-effects-on-interview-speed-related-to-interviewer-3VIdnNDqkM
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
2325-0984
eISSN
2325-0992
D.O.I.
10.1093/jssam/smx034
Publisher site
See Article on Publisher Site

Abstract

Abstract It can be expected that interview speed and straight-lining tendency are related: a higher speed can lead to an increase in cognitive difficulty, which might be coped with by straight-lining; whereas, straight-lining can decrease response latency and hence increase the interview speed. Previous research shows that interview speed is influenced by the interviewer. Straight-lining tendency and other satisficing symptoms have also been shown to be subject to interviewer effects, albeit to a smaller extent. In the first step in this paper, we analyze interviewer effects on interview speed and straight-lining in one module of the European Social Survey, round 7. We then study the correlation between interview speed and straight-lining, without determining causality, and separate this into an interviewer-level and respondent-level correlation. The results confirm previous findings: a large intra-interviewer correlation coefficient for interview speed and a smaller intra-interviewer correlation coefficient for straight-lining. Further, the results show a significant positive correlation in nine out of fifteen countries at the respondent level and in five (different) countries at the interviewer level. The latter indicates that “fast” interviewers are the ones carrying out interviews during which more straight-lining occurs. 1. INTRODUCTION Research into the relationship between response time and straight-lining–two indicators of satisficing (Krosnick 1991; Krosnick, Narayan, and Smith 1996)–has mainly been carried out for online surveys, possibly because response times are easily collected for that mode (Heerwegh 2003) and because straight-lining is more likely to occur in self-administered surveys than in telephone or face-to-face equivalents (e.g., Fricker, Galesic, Tourangeau, and Yan 2005). As a consequence, existing research on this relationship is also limited to the effects of respondent and/or item characteristics. Studies show that speeding (high speed, short response time) is related to straight-lining (Conrad, Tourangeau, Couper, and Zhang 2011; Zhang and Conrad 2014; Revilla and Ochoa 2015). However, the relationship between response time and data quality in general is ambiguous. On the one hand, research generally shows that speeding is related not only to straight-lining but also to primacy effects (Malhotra 2008), less-detailed answers (Wells, Rao, Link, and Pierce 2012), and incoherence in answers (Revilla and Ochoa 2015). On the other hand, long response times have been linked to ambivalent attitudes (Johnson 2004), lack of knowledge (Heerwegh 2003), usability problems (McClamroch 2011), and, by contrast, engaged respondents (Crawford, Couper, and Lamias 2001). Response times, often referred to as response latencies, are a measurement of the cognitive effort made by the respondent to understand and answer a question (Bassili and Scott 1996). Hence, short response times can be a sign of satisficing or of a cognitive shortcut taken by the respondent (Krosnick 1991). However, short response times can also reflect a respondent’s accessible attitude (Grant et al. 2010), knowledge (Heerwegh 2003), or higher cognitive abilities (Bassili 1996; Yan and Tourangeau 2008; Couper and Kreuter 2013). However, Olson and Bilgen (2011) found longer interview durations for higher-educated respondents. This could be the consequence of more engagement, or of less satisficing, when completing a survey. Hence, respondent characteristics—in particular, cognitive abilities and motivation—together with item characteristics can influence response times in different directions. The relationship between speeding and data quality has also been found to be dependent on the respondent’s education level (Malhotra 2008; Zhang and Conrad 2014). In face-to-face surveys, the speed of an interview is jointly affected by the interviewer and the respondent. This makes the relationship between speed and data quality even more complex. Interviewers are responsible for the reading speed of the questions, the time given to respondents to answer, and the length of breaks. Respondents influence the interview speed through response latency. Based on the principle of standardized interviewing (Fowler and Mangione 1990), there is no reason why between-interviewer variation in interview speed should occur after controlling for proxies of a respondent’s cognitive capacity, such as age and education. In other words, after acknowledging that interviewers should adapt their speed to the respondents’ cognitive capacity, we should not expect differences between interviewers. However, research shows that in face-to-face surveys, interviewers do play a role in the interview speed (Hox 1994; Japec 2005; Olson and Peytchev 2007; Loosveldt and Beullens 2013a, 2013b). Cannell, Miller, and Oksenberg (1981, pp. 414–415) observed that the interviewer’s major goal seems to be to finish the interview as quickly as possible. This observation suggests that the interviewer can deliberately increase the interview speed. This tendency can vary between interviewers, depending upon, for instance, their motivation and workload. The tendency of interviewers to increase the interview speed could be an indication of their burden: interviewers having too many interviews to conduct or a lack of motivation to do their job properly (West and Blom 2016, p. 19). Olson and Peytchev (2007) found that interviewers with more experience perform their interviews faster and that the speed also increases over the fieldwork period. One can hypothesize that the questionnaire becomes familiar to the interviewer, or that the training in standardized interviewing focuses more on new interviewers than on the more experienced ones. Using data from the European Social Survey, Loosveldt and Beullens (2013b) confirmed these results and additionally found indications that the interview speed is related to the type and the complexity of the items. In summary, all the existing results of analyses of interview speed support the idea that interviewers can influence how fast the interview proceeds as much as respondents can. Literature about interviewer effects on straight-lining is sparser. Loosveldt and Beullens (2017) found that variation in the straight-lining behavior of respondents can be partly explained by interviewer effects, even after controlling for respondent characteristics (age, gender, and education level), the rank of the interview for the same interviewer (whether it was the first, second, third, etc., carried out by the interviewer during the data collection period), density (self-reported indication of population density), region, and the interviewer assessment of how well the respondent performed during the interview. Their results also show that in general—similarly to the interview speed—straight-lining increases with the rank of the interview. Further, interviewer effects have been shown to have an influence on other indicators of satisficing. Research shows that interviewers have an effect on acquiescence (Hox, De Leeuw, and Kreft 1991; Olson and Bilgen 2011). Olson and Bilgen (2011) found more acquiescence behavior among respondents who completed an interview with an experienced interviewer, but no relationship with interview speed. Further, they did not find any link between acquiescence and the gender, age, or education level of interviewers. They concluded that interviewer effects on acquiescence are caused by interviewer behavior: one can assume that more experienced interviewers establish a good relationship with respondents, which can lead to more acquiescence. Pickery and Loosveldt (2004) also found significant interviewer effects on the number of “don’t know” and “no opinion” answers. Lastly, Turner’s findings (2013; Turner, Sturgis, and Martin 2015) suggest that interviewers with a less accepting attitude to respondent refusal elicit more satisficing behavior among their respondents, whereas no links with interviewers’ demographic characteristics, skills, or personalities were found. Based on all these findings, a further analysis of straight-lining, as one indicator of satisficing, and its relationship to interview speed in face-to-face surveys seems relevant and appropriate. To the best of our knowledge, no previous research has been performed on how interviewers can affect the relationship between interview speed and straight-lining tendency in face-to-face surveys. This is the main research objective of the current paper. 2. RESEARCH OBJECTIVES In the first step, we separately study interviewer effects on interview speed (RQ1) and straight-lining (RQ2) in round 7 of the European Social Survey (ESS7). Although we expect to replicate previous findings regarding respondent and interviewer effects on interview speed, we consider this a necessary first step in order to validate our data and to underpin the importance of the further steps undertaken. It is also a relevant additional test of the external validity of previous results (Olson and Peytchev 2007; Loosveldt and Beullens 2013b). Next, we determine whether there is a relationship between interview speed and straight-lining tendency in face-to-face surveys (RQ3a). We do not hypothesize any causality in this regard, but expect these aspects of interviewer and response behavior to be positively correlated. Indeed, these two interview characteristics can be regarded as indications of interviewer and/or respondent satisficing behavior. Moreover, if a high interview speed is largely defined by the interviewer, this may increase the difficulty of answering the questions, and hence, trigger more satisficing in the respondent’s behavior, leading to straight-lining tendency, among other things. In reverse, straight-lining—which can be elicited by interviewer behavior—might increase the interview speed, with respondents not going through the full cognitive response process and answering faster. Accordingly, we consider that the interviewer behavior and the respondent behavior might both be responsible for the expected positive correlation between speed and straight-lining. Therefore, we also analyze how the correlation between interview speed and straight-lining tendency can be separated into the respondent level and the interviewer level (RQ3b). If a part of the correlation is attributed to the interviewer level, this would suggest that interviewer satisficing occurs, with some interviewers leading the interviews at a high speed and/or not motivating the respondents to give the best possible answers. For the three research objectives presented above, we also discuss possible differences between the countries participating in the ESS7. Overall, we do not expect differences between countries when they properly implement the survey according to the specifications of standardized interviewing. The comparison of different countries can, however, be considered an additional test of external validity. More information about the ESS7 is provided in the next section. 3. DATA: THE SEVENTH ROUND OF THE EUROPEAN SOCIAL SURVEY To answer our research questions, we use data from the European Social Survey round 7 (ESS 2014).1 Fifteen countries participated in the ESS7 using computer-assisted personal interviews (CAPI) (Austria, Belgium, Denmark, Estonia, Finland, France, Germany, Hungary, Ireland, the Netherlands, Norway, Slovenia, Sweden, Switzerland, and the United Kingdom) and six using (at least partly) paper-assisted interviews (the Czech Republic, Israel, Lithuania, Poland, Portugal, and Spain). We exclude the latter group from our analyses, as the recording of start and end times suffers from notable errors in this type of interview. In all the countries, samples were randomly selected—although sampling designs can vary (ESS 2015)—and should reach an effective sample size of 1,500 individuals. In addition to its goal of measuring changes in attitudes and beliefs in Europe, the ESS aims to reach the highest methodological standards to ensure valid cross-country comparison. Therefore, among many methodological projects conducted by the ESS, all participating countries using CAPI have had timers in their software since the ESS round 5. The timers measure the start (just before the first item, A1) and the end (immediately after the last item, F61) of the interview and also the start and end of each module (thematic group of items) in the questionnaire. This is the most detailed measurement of duration/speed available in the ESS round 7; no time data is available at the item level. The ESS questionnaire contains four core modules, largely similar from round to round, and two rotating modules. In round 7, immigration and health were the topics of the rotating modules. In the current analysis, we concentrate on the mean speed over Module B, the second module of the ESS core questionnaire, which covers political involvement, including political interest, trust, electoral, and other forms of participation; party allegiance; and socio-political orientation. We chose this module because we are interested in straight-lining in blocks or grids of more than three items. There are four such blocks in the whole round 7 questionnaire, three of which are in Module B, and the items in these blocks represent more than one-third of those in this module. Restricting the analysis to one module ensures that the time measurement is more closely related to the answering process for the items used to measure straight-lining. 4. MEASURING AND MODELING INTERVIEW SPEED: INTERVIEWER AND RESPONDENT EFFECTS The interview speed is defined as the number of items answered per minute. To calculate the mean interview speed over Module B, we use a variable that indicates the duration of the module in minutes (BINWTM). The recorded module interview length differs between countries, but ranges from 0 to 311 minutes. We remove extreme values from the analysis, defined as less than 5 minutes and more than 40 minutes. Loosveldt and Beullens (2013b) showed that language has a low impact on the interview speed and, hence, that the extreme values can be defined across countries. Moreover, we believe that the defined extreme values are unlikely, independent of the language, and must be due to technical or procedural issues. Lastly, the percentage of missing values before excluding extreme values varies between countries, from 0.8% (Finland) to 10.2% (the Netherlands). In the second step, we count the number of items in Module B that were presented to each respondent. The ESS questionnaire contains filter items, and as a consequence, the number of items asked can vary from one respondent to another. The presence of filters in Module B is limited and the difference between the minimum and maximum number of items is small (a minimum of 40 and a maximum of 44). Lastly, we calculate the mean module speed as the number of items asked in Module B, q, divided by the Module B length t in minutes for each respondent i: vi=qiti. Table 1 shows the mean Module B speed and the standard deviation over all respondents and for each country, after having removed the extreme values. The mean speed ranges from 3.22 items per minutes in Germany to 5.25 in Ireland. The mean speed we obtain is slightly lower than in the results obtained by Loosveldt and Beullens (2013b, p. 1425): three to four items per minutes compared with four to five items per minute. This could be due to different treatment of extreme values. Table 1. Descriptive Statistics of the Interview Speed in Module B Per Country Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 Table 1. Descriptive Statistics of the Interview Speed in Module B Per Country Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 Country n Mean speed Standard deviation Austria 1,663 4.00 1.60 Belgium 1,690 4.17 1.29 Denmark 1,477 3.89 1.27 Estonia 1,874 4.12 1.67 Finland 2,071 4.25 1.29 France 1,830 3.69 1.10 Germany 2,977 3.22 1.09 Hungary 1,670 4.23 1.52 Ireland 2,285 5.25 1.54 The Netherlands 1,723 3.87 1.31 Norway 1,399 3.97 1.31 Slovenia 1,163 4.48 1.44 Sweden 1,753 3.57 1.10 Switzerland 1,501 4.02 1.39 United Kingdom 2,182 4.59 1.46 To model the module speed and to evaluate the interviewer effects, we use a multilevel model in which respondents are clustered in interviewers. For each country separately, we assume that the module speed is not only determined by the respondents—as it would be for self-administered questionnaires—but also by the interviewers and that part of the variability can be explained by the differences between the interviewers at the second level. Because respondents are not randomly assigned to interviewers, but instead to interviewers who work in their vicinity, interviewer effects (intra-interviewer correlation) can be confounded with area effects (e.g., O’Muircheartaigh and Campanelli 1998; Schnell and Kreuter 2005; Vassallo, Durrant, and Smith 2017). The intertwining of interviewer and area effects is a complex problem, which can be partly circumvented by controlling for some relevant respondent characteristics. As already mentioned, existing research shows that a respondent’s age and level of education, which can serve as a proxy for cognitive abilities, influence the interview speed, although results are not consistent with regard to the direction of this relationship (e.g., Olson and Bilgen 2011; Loosveldt and Beullens 2013a, 2013b). It has also been shown that the rank of the interview plays a role in the interview speed (Olson and Peytchev 2007; Loosveldt and Beullens 2013a). Hence, we control for these characteristics, as well as for these characteristics: respondent’s gender; whether or not the interview was conducted in the respondent’s mother tongue2 (Lang); the population density as described by the respondent on a five-point scale (Dens); the region based on counties, provinces, or other subnational entities (Reg); the interviewer assessment on a five-point scale of whether the respondent answered to the best of their abilities (Abil); and of whether the respondent understood the questions (Und). These are characteristics that we believe may influence the interview speed. The following model (Model A) is a random-intercept, fixed-slopes design in which we control for respondent and interview characteristics: vi,j=β0,j+β1Womani,j+β2Langi,j+β3Educ1i,j+β4Educ2i,j+β5Agei,j+β6Ranki,j+β7Densi,j+β8Regi,j+β9Abili,j+β10Undi,j+ɛi,j, with β0,j=γ0,0+u0,jwhere β0,j is the random intercept, βk from k = 1 to k = 6, are the fixed effects, γ0,0 ⁠, the overall intercept; ɛi,j∼N0,σw2; and u0,j∼N(0,σb2) ⁠. The intra-interviewer correlation coefficient ρv that represents the portion of variance in speed that is explained by the variation between interviewers is given by: ρv=σb2σw2+σb2. For each country for Module B in the ESS7, table 2 displays the overall intercept γ0,0 ⁠, the fixed slope for the dependent variables gender β1 ⁠, whether or not the interview was taken in the respondent’s mother tongue (Lang) β2 ⁠, level of education (upper-secondary β3 and tertiary β4 ⁠), age β5 ⁠, rank β6 ⁠, population density (Dens) β7 ⁠, whether the respondent answered to the best of their ability (Abil) β9 ⁠, whether the respondent understood the questions (Und) β10 ⁠, the between variance σb2 and the residual variance σw2 ⁠, and the intra-class correlation coefficient ρv ⁠. There are a number of categorical variables: “Woman” (0 = man, 1 = woman); “Lang” (0 = language of interview is not mother tongue, 1 = language of interview is mother tongue); two dummies for education based on ISCED level, “Educ1” (1 = ISCED level 3 or 4, 0 = other); and “Educ2” (1 = ISCED level 5, tertiary, 0 = other). Age and rank are continuous variables. The population density, the respondent abilities, and the respondent’s understanding are considered continuous. We do not display the effects of the region; these are lengthy to report and often not significant at the 0.05 level, except for a few regions in Belgium, France, Germany, Hungary, Ireland, and Norway. All the codes for data preparation and modelling can be found in the supplementary materials. Table 2. Parameter Estimates of Model A for Module Speed (ESS7) Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 * p < 0.10, **p < 0.05. Table 2. Parameter Estimates of Model A for Module Speed (ESS7) Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 Country nr ni γ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und σb2 σw2 ρv Austria 1,663 88 4.07 −0.04 −0.07 0.01 0.14 −0.00* 0.01* −0.07 −0.06 0.05 0.26 2.23 0.10 Belgium 1,690 150 6.56 −0.04 0.63** 0.08 0.08 −0.02** 0.03** 0.04 −0.17** 0.31** 0.22 1.13 0.14 Denmark 1,465 88 3.02 −0.11* 0.49** 0.06 −0.05 −0.02** 0.01** −0.03 −0.02 0.28** 0.26 1.21 0.16 Estonia 1,833 140 4.15 0.01 0.17 −0.08 0.04 −0.02** 0.02** −0.06 −0.08** 0.21** 0.72 1.73 0.26 Finland 2,058 137 4.63 −0.10** 0.94** 0.17** 0.22** −0.03** 0.01 −0.01 −0.16** 0.41** 0.14 1.14 0.08 France 1,824 132 2.87 0.03 0.37** −0.01 0.02 −0.02** 0.02** 0.02 −0.04 0.20** 0.24 0.80 0.20 Germany 2,945 266 3.67 −0.14** 0.38** 0.02 0.10 −0.02** −0.00 0.05* 0.00 0.20** 0.24 0.83 0.20 Hungary 1,584 137 4.23 −0.01 0.16 −0.06 0.13 −0.00* 0.02** 0.05 −0.05 0.03 0.68 1.39 0.30 Ireland 2,273 112 4.66 −0.03 0.50** −0.09 −0.18** −0.01** 0.02** 0.04 −0.01 0.08** 0.69 1.70 0.28 The Netherlands 1,717 113 4.86 0.01 0.28** −0.03 −0.06 −0.02** 0.01** −0.03 −0.02 0.31** 0.34 1.14 0.20 Norway 1,315 64 3.42 −0.31** 0.74** 0.03 0.02 −0.02** 0.01** −0.10** −0.08 0.47** 0.15 1.18 0.09 Slovenia 1,158 60 6.14 −0.02 −0.26 0.07 0.05 −0.03** 0.02** 0.04 −0.10 −0.02 0.50 1.29 0.19 Sweden 1,750 91 2.42 −0.10** 0.39** −0.06 0.08 −0.02** 0.01** −0.01 0.03 0.29** 0.18 0.85 0.15 Switzerland 1,497 64 2.73 −0.13** 0.00 0.03 0.22** −0.01** 0.01** 0.02 −0.02 0.32** 0.67 0.99 0.37 United Kingdom 2,069 210 4.14 −0.15** 0.71** −0.05 −0.10 −0.03** 0.02** 0.05 −0.09 0.35** 0.33 1.44 0.16 * p < 0.10, **p < 0.05. The sample sizes vary from 1,158 (after removing missing durations and extreme values) in Slovenia, with the lowest number of interviewers (60), to 2,945 in Germany, with the highest number of interviewers (266). The average number of respondents per interviewer (i.e., the workload) varies from 9.9 in the United Kingdom to 23.4 in Switzerland. The overall intercept also varies substantially between countries, ranging from 2.42 items per minute in Sweden to 6.14 items per minute in Slovenia. In regard to RQ1, the intra-interviewer correlation coefficients are all above 8% (Finland), with a maximum value of 37% (Switzerland). Therefore, interviewer effects on speed cannot be neglected, regardless of the country. Indeed, Groves (1989, p. 364) notes: “Most other surveys have an average ρint between 0.01 and 0.02.” Although acceptable values for the intra-interviewer coefficient depend on the number of interviewers, the sample size, and interaction with area effects, we consider interviewer effects as substantial whenever the intra-interviewer correlation coefficient exceeds 0.05. For comparison, in round 7 the intra-interviewer correlation coefficient for the respondent age varies between 0.00 in Belgium and 0.06 in Estonia. These findings confirm our expectations based on previous research (Hox 1994; Japec 2005; Olson and Peytchev 2007; Loosveldt and Beullens 2013b). Moreover, the results underline the importance of analyzing the link between the effect of interviewers on interview speed and indicators of data quality such as straight-lining (RQ3b). Although not the core interest in this paper, we briefly discuss the effects of the independent variables in Model A. Women have slower interviews in Finland, Germany, Norway, Sweden, Switzerland, and the United Kingdom. When the interview language is the respondent’s mother tongue, the interview speed is mostly higher, as would be expected (the exceptions are Austria and Slovenia, but non-significant), although the effect is not consistently significant. This might, however, depend on the size of the group for which the mother tongue is not the interview language. The direction of the effect of education is not clear, and the effect itself is quite often not significant. This could be the result of a mixed effect of higher-educated people having greater cognitive abilities, but also using more cognitive effort to give an answer. Having an upper-secondary (Educ1) and/or tertiary (Educ2) level of education significantly increases the speed, compared with lower-secondary education or below in two countries (Finland and Switzerland) and significant decreases in speed in Ireland. However, the pattern is not necessarily linear; compared with people having a lower education level, the increase in speed for people with a tertiary level of education does not always exceed that for people with an upper-secondary level of education. The interview speed also decreases with the respondent’s age in all countries (Austria not significant) and increases with the rank of the interview (Austria, Finland, and Germany not significant). This supports the idea that there is a “learning” effect for interviewers, which is in line with previous findings (Olson and Peytchev 2007; Loosveldt and Beullens 2013b). The population density almost never has a significant effect on the interview speed, except for Norway where people in less dense areas have slower interviews. The interview speed decreases when the respondent more frequently answers to the best of their ability, a proxy for motivation (only significant in Belgium, Estonia, and Finland), whilst the speed increases when the respondent more frequently understands the questions. However, these two variables are reported by the interviewers and could therefore also suffer from interviewer effects. The model (Model A) was also run without the independent variables at the respondent level. The resulting proportions of variance explained by the variance between interviewers, without controlling for respondent characteristics, are close to those shown for Model A. 5. MEASURING AND MODELING STRAIGHT-LINING TENDENCY: INTERVIEWER AND RESPONDENT EFFECTS Although the definition of straight-lining is clear—the tendency to give the same answer to items regardless of the content of the item—relevant literature suggests different possible ways to measure this (Mulligan, Krosnick, Smith, Green, and Bizer 2001; Chang and Krosnick 2009; Zhang and Conrad 2014; Loosveldt and Beullens 2017). The first point for discussion concerns the use of a homogeneous or a heterogeneous set of items. A heterogeneous set of items may be seen as better suited because it might reasonably be expected that the respondent’s answer will differ from one item to another if the items have a low inter-correlation. Giving the same answer to a homogenous set of items that are nuances of the same topic may reflect true values, rather than being caused by straight-lining. However, in the current analysis we are limited to homogeneous sets of items, because the ESS questionnaire does not contain a suitable set of heterogeneous items. The second discussion point concerns pure straight-lining versus straight-lining tendency. Zhang and Conrad (2014) studied pure patterns in an online survey, where exactly the same answer was given for all the items in a grid. We argue, however, that this is more likely to occur in an online questionnaire than in a face-to-face interview and more likely to occur for shorter sets of items than for longer sets. The choice to only consider pure straight-lining does not shed any light on straight-lining tendencies, which are also important and which we expect to find in face-to-face surveys with longer sets of items. We therefore measure the straight-lining tendency by examining the percentage of answers that are exactly the same as the previous one in a set, including refusals, “don’t know,” and “no answer” (Loosveldt and Beullens 2017). For example, examining the following answers to a set of two times four items which are considered as a whole: 3, 3, 2, 1 and 1, 1, 4, 2, the percentage of straight-lined answers is 33.3% (two out of six, with one identical subsequent answer in each set, ignoring the transition between sets). There is a potential maximum of three straight-lined answers in each set, as the first item obviously has no previous answer to compare it with. The measurement can, in this example, only take seven values (0%, 16.7%, 33.3%, 50%, 66.6%, 83.3%, or 100%). Hence 33.3% may seem large, although looking at the answers given in the example, straight-lining is not obvious. The measurement is better suited and more robust for more blocks and more items in blocks (as in the ESS7 Module B). Other ways to measure straight-lining could, however, be considered. One possibility would be to examine the standard deviation of the scores for each block and each respondent. A low standard deviation would be a sign of low discrimination or non-discrimination in the answers in one block. However, a switching pattern—for instance, 2, 3, 2, 3 and 4, 5, 4, 5—would also result in a low standard deviation, although these answers do not follow a straight-lining pattern. Another possibility would be to use Mulligan’s score (Mulligan et al. 2001; Chang and Krosnick 2009). Mulligan’s score is a distance metric, measuring the average square root of the absolute difference between any two answers by the same respondent in a block of items, or n2-1∑i=1n∑i'>inxi-xi' ⁠, where n is the number of items in the grid and x is the answer of the respondent to the item. The standard deviation method and Mulligan’s score are, however, distance measurements that evaluate non-discrimination or low discrimination and do not reflect the sequence pattern. Moreover, based on the sensitivity analysis detailed by Loosveldt and Beullens (2017), we believe that the results are in practice independent of the chosen measurement. Table 3 presents the three blocks or sets of items and their characteristics that we use to measure straight-lining tendency. We chose blocks that contain more than three items, and these all contain 11-point scale items, which does not allow us to study scale effects. The total number of items in these blocks is seventeen, which represents more than one-third of all the items in Module B. The maximum number of straight-lined answers is fourteen (pure straight-lining). This means that the measurement of the tendency for straight-lining can take fifteen values (0%, 100*1/14%, 100*2/14% … 100*13/14%, 100%). Table 3. Blocks of Items Used in the Calculation of the Percentage of Straight-Lined Answers Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Table 3. Blocks of Items Used in the Calculation of the Percentage of Straight-Lined Answers Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Block No. of items scale B1a–B1f: relation to politics 6 11-point scale (0–10) B2–B8: trust in institutions 7 11-point scale (0–10) B20–B25: satisfaction 4 11-point scale (0–10) Table 4 displays, for each country, the mean straight-lining tendency measurement and the standard deviation over all respondents. The mean straight-lining tendency ranges from 19.8% (slightly less than three straight-lined items) in Germany to 28.6% (between four and five straight-lined items) in Hungary. Table 4. Descriptive Statistics of the Straight-Lining Tendency Measurement in Module B Per Country Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 Table 4. Descriptive Statistics of the Straight-Lining Tendency Measurement in Module B Per Country Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 Country n Mean Standard deviation Austria 1,663 24.6 14.9 Belgium 1,692 23.4 13.0 Denmark 1,477 20.8 11.2 Estonia 1,874 26.1 14.0 Finland 2,071 22.6 12.5 France 1,830 22.3 12.6 Germany 2,977 19.8 11.5 Hungary 1,670 28.6 15.0 Ireland 2,285 23.6 13.5 The Netherlands 1,723 22.2 11.9 Norway 1,399 20.8 11.0 Slovenia 1,163 25.7 14.4 Sweden 1,753 20.1 11.3 Switzerland 1,501 22.0 11.7 United Kingdom 2,182 22.6 12.5 In a similar way to the interview speed, interviewer effects on straight-lining can be modeled by a multilevel model to separate respondent variance from interviewer variance. We control for the same respondent characteristics as the ones we use in relation to the interview speed: age and level of education (as proxies for cognitive abilities), the rank of the interview, gender, whether or not the interview was in the respondent’s mother tongue, population density, region, whether the respondent answered to the best of their ability, and whether the respondent understood the questions. The following model (Model B) is a random-intercept, fixed-slopes model in which we control for respondent and interview characteristics: stri,j=ϕ0,j+ϕ1Womani,j+ϕ2Langi,j+ϕ3Educ1i,j+ϕ4Educ2i,j+ϕ5Agei,j+ϕ6Ranki,j+ϕ7Densi,j+ϕ8Regioni,j+ϕ9Abili,j+ϕ10Undi,j+ωi,j, with ϕ0,j=μ0,0+v0,j ⁠.where ϕ0,j is the random intercept, ϕk from k = 1 to k = 6 are the fixed effects, μ0,0the overall intercept, ωi,j∼N(0,τw2) ⁠, and v0,j∼N(0,τb2) ⁠. The intra-interviewer correlation coefficient ρstr ⁠, which represents the proportion of variance in straight-lining tendency that is explained by variation between interviewers is given by: ρstr=τb2τw2+τb2 Table 5 displays, for each country, the overall intercept μ0,0 ⁠; the fixed slope for the dependent variables; gender ϕ1 ⁠; whether the interview was in the respondent’s mother tongue ϕ2 ⁠; level of education ϕ3; ϕ4 ⁠; age ϕ5 ⁠; rank ϕ6 ⁠; population density ϕ7 ⁠; whether the respondent answered to the best of their ability ϕ9 ⁠; whether the respondent understood the questions ϕ10 ⁠; the between variance τb2 and the residual variance τw2 ⁠; and the intra-class correlation coefficient ρstr for Module B in the ESS7. Again, to stay concise the region effect is not displayed, but is significant for a few regions in Belgium, France, Hungary, Germany, Ireland, Slovenia, and Sweden. Table 5. Parameter Estimates of Model B2 for Straight-Lining Tendency (ESS7) Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 * p < 0.10, **p < 0.05. Table 5. Parameter Estimates of Model B2 for Straight-Lining Tendency (ESS7) Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 Country nr ni μ0,0 Woman Lang Educ1 Educ2 Age Rank Dens Abil Und τb2 τw2 ρstr Austria 1,784 88 37.50 1.69** −4.83** −1.15 −3.08** 0.01 0.02 −0.54 −0.25 −1.39* 39.90 163.17 0.19 Belgium 1,767 150 26.30 1.52** −1.64 −1.17* −2.72** 0.07** 0.16** 0.49 −1.28 −0.14 8.01 14.51 0.05 Denmark 1,490 88 17.35 −0.01 −0.65 −1.06 −0.46 0.08** 0.01 −0.49 −0.32 0.64 3.14 117.04 0.03 Estonia 2,006 141 36.88 0.21 1.29 −1.34 −3.26** 0.06** −0.00 0.13 −1.06 −2.11** 26.73 164.00 0.13 Finland 2,074 137 19.86 1.65** −0.54 −1.51** −1.56** 0.03** 0.07 0.19 −0.07 −1.34** 0.17 113.21 0.00 France 1,910 132 35.23 0.81 −1.57 0.71 −2.87** 0.06** 0.08* 0.44* −2.80** −0.68 1.40 146.11 0.01 Germany 3,012 267 16.25 1.86** −2.30** 0.24 −1.08 0.06** 0.01 0.46* 0.13 −0.77** 5.39 119.36 0.04 Hungary 1,612 137 39.20 1.59** −4.38 −0.43 −2.86** 0.01 0.08 0.10 −1.25** −1.44** 48.91 163.81 0.21 Ireland 2,378 112 33.78 1.45** −1.68 −1.09* −3.84** −0.00 0.04 −0.52* 0.08 −1.72** 36.08 138.85 0.20 The Netherlands 1,913 113 32.38 0.65 −1.08 −1.27* −0.66 0.03* −0.04 −0.30 0.05 −1.86** 4.20 132.63 0.03 Norway 1,345 64 26.86 0.00 −2.56** 0.38 0.42 0.07** 0.05 −0.12 −1.37** −0.08 0.63 117.04 0.01 Slovenia 1,218 60 39.65 1.61** −8.75** −0.58 −3.42** 0.03 0.11** 0.77* −1.26* −1.75** 6.28 188.81 0.03 Sweden 1,788 91 16.92 1.33** −2.10** 0.06 0.09 0.06** 0.05* −0.04 −0.13 −0.19 3.45 119.44 0.03 Switzerland 1,528 64 29.60 0.45 1.43* −0.73 −1.86* 0.00 0.03 0.00 −0.02 −1.58** 3.08 131.06 0.02 United Kingdom 2,142 210 39.72 2.06** −0.73 −1.70** −2.62** −0.04** 0.00 0.17 −1.16** −2.36** 3.40 146.50 0.02 * p < 0.10, **p < 0.05. The overall intercept varies from 16.25% of straight-lined answers in Germany (between two and three items out of the fourteen) to 39.20% in Hungary (close to six items out of the fourteen). In countries in which gender has a significant effect at the 0.05 level (Austria, Belgium, Finland, Germany, Hungary, Ireland, Slovenia, Sweden, and the United Kingdom), women tend to straight-line more often. Taking the interview in the mother tongue reduces the percentage of straight-lined answers in Austria, Germany, Norway, Slovenia, and Sweden. Generally, the effect of (higher) education is negative (a lower straight-lining tendency). Further, the percentage of straight-lined answers increases with age (with the exception of the United Kingdom), and the rank of the interview has no effect on the tendency to straight-line (with the exception of Belgium). The population density also does not have an effect on straight-lining. Both higher respondent motivation (Abil) and understanding (Und) lead to fewer straight-lined answers. The intra-class correlations (ICCs) are lower than 0.05 in Denmark, Finland, France, Germany, the Netherlands, Norway, Slovenia, Sweden, Switzerland, and the United Kingdom, and lower than 0.10 in Belgium. In Austria, Estonia, Hungary, and Ireland, the ICCs are above 0.13. These differences between countries are in line with previous findings (Loosveldt and Beullens 2017). Moreover, the results suggest that the influence of the interviewer on the straight-lining behavior of the respondent is smaller than the influence of the interviewer on the interview speed. This is not surprising, since the interviewer impact on straight-lining is less direct (through motivation and task difficulty) than on interview speed, in which the interviewer plays an active role. Again, controlling for respondent characteristics has only a small effect on the proportion of variance explained at the interviewer level, suggesting that the interviewer effects are larger than the clustering effects due to homogenous groups of respondents in particular areas. 6. SEPARATING THE CORRELATION BETWEEN INTERVIEW SPEED AND STRAIGHT-LINING TENDENCY INTO THE INTERVIEWER LEVEL AND THE RESPONDENT LEVEL Our main research questions (RQ3a and RQ3b) concern the correlation between straight-lining tendency and interview speed, and the effect of interviewers on this correlation. Accordingly, we are interested not only in separating interview speed and straight-lining tendency variances into the respondent level and the interviewer level, but also in the effect of the respondents and the interviewers on the relationship between these two proxies of satisficing. The correlations within countries at the respondent level—without taking into account the nesting within interviewers (RQ3a)—between interview speed and straight-lining tendency are presented in table 6 (column correlations). Eight of the countries display a small but significant correlation (not exceeding 0.13) between straight-lining tendency and interview speed, suggesting that such a relationship may exist. Moreover, most correlations are—as expected—positive, indicating that a faster interview speed is related to an increased straight-lining tendency. We can hypothesize different reasons for this low correlation, as the interaction between the interviewer and the respondent is complex. First, the speed is not measured specifically over the items in blocks, but over the whole module, which may reduce the amplitude of the correlation. Moreover, the interviewer and respondent both affect the speed and the straight-lining behavior, but the interviewer effects on speed are larger and more straightforward (a direct role through reading speed). Lastly, the relationship of speed to data quality is ambiguous. On the one hand, a high speed can be a sign of satisficing and can lead, for instance, to straight-lining; but it can also be a sign of the respondent’s knowledge and understanding. On the other hand, a lower speed can be a sign of (too high) cognitive difficulty, but also of more cognitive effort to provide an answer. Finland has a negative significant correction between interview speed and straight-lining. This is an unexpected result. Although we do not have recordings that would be necessary to understand this phenomenon, we can hypothesize that straight-lining is a satisficing behavior due to the cognitive difficulty the respondent experiences and that the interviewer slows down the interview to help the respondent. Table 6. Correlations between Straight-Lining Tendency and Interview Speed and the Separation into Within and Between Components Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* p < 0.10 and *p < 0.05 Table 6. Correlations between Straight-Lining Tendency and Interview Speed and the Separation into Within and Between Components Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* Country ρstr ρv Correlations Corr. within Corr. between Austria 0.19 0.10 0.02 0.02 0.10 Belgium 0.05 0.14 0.05* 0.08* 0.14 Denmark 0.03 0.16 0.05 0.08* 0.22 Estonia 0.12 0.26 0.12* 0.04 0.56* Finland 0.00 0.08 −0.05* 0.06 0.06 France 0.01 0.20 0.03 0.07* 0.40 Germany 0.04 0.20 0.03 0.03‘ 0.15 Hungary 0.21 0.30 0.04 0.03 −0.01 Ireland 0.20 0.28 0.13* 0.05* 0.51* The Netherlands 0.03 0.20 0.05* 0.05* 0.15 Norway 0.01 0.09 −0.03 0.03 0.21 Slovenia 0.03 0.19 0.13* 0.16* 0.43* Sweden 0.03 0.15 0.00 0.05* −0.01 Switzerland 0.02 0.37 0.10* 0.07* 0.44* United Kingdom 0.03 0.16 0.06* 0.05* 0.43* p < 0.10 and *p < 0.05 In the next step we specify a multivariate multilevel model (Heck and Thomas 2015, Chapter 6). With this model, we can separate the relationship between interview speed and straight-lining tendency into the interviewer and the respondent level without specifying the direction of this relationship. Therefore, we specify both interview speed and straight-lining tendency as dependent variables. To estimate the variance and covariance between interview speed and straight-lining tendency at both levels, we consider the following multivariate model, controlling for respondent characteristics (Models A and B) as illustrated in figure 1. Figure 1. View largeDownload slide Multivariate multilevel model with explanatory variables at the respondent level. Figure 1. View largeDownload slide Multivariate multilevel model with explanatory variables at the respondent level. We then estimate the following model: v(i,j)stri,j=β0,jϕ0,j+00β00ϕv(i,j)str(i,j)Xi,j+ɛi,jωi,j β0,jϕ0,j=γ0,0μ0,0+u0,jv0,j with covariance matrices ΨW=σw2ϑv,strϑv,strτw2 at the within level and ΨB=σb2αv,strαv,strτb2 at the between level. Conforming to models A and B, the random intercepts for the interview speed and straight-lining tendency are given by respectively β0,j and ϕ0,j ⁠, and the overall intercepts are given by γ0,0 and μ0,0 ⁠. The regression coefficient vectors for interview speed and straight-lining tendency are denoted by β=(β1β2β3β4β5β6β7β8β9β10) and ϕ=(ϕ1ϕ2ϕ3ϕ4ϕ5ϕ6ϕ7ϕ8ϕ9ϕ10) ⁠. The symbol Xi,j denotes the vector of respondent characteristics for which we control. The respondent-level variances are given by σw2 and τw2, and the interviewer-level variances are given by σb2 and τb2 ⁠. The main parameters we are interested in are the covariances between interview speed and straight-lining tendency at the respondent level (within) ϑv,str and at the interviewer level (between) αv,str ⁠. Correlations are calculated using standardized covariances. The resulting correlations between straight-lining tendency and interview speed separated into the interviewer and the respondent level, after controlling for respondent characteristics, are displayed in the two last columns of table 6 (Corr. within and Corr. between). The results of the separation into the respondent level (within) and the interviewer level (between) of the correlation, addressing our last research question (RQ3b), are interesting. In nine countries (Belgium, Denmark, France, Ireland, the Netherlands, Slovenia, Sweden, Switzerland, and the United Kingdom), a small but significant positive correlation is found at the respondent level between straight-lining tendency and speed. This indicates a positive relationship between speed and straight-lining independent of the interviewer, probably meaning that these are both indicators of respondent satisficing. The correlation between speed and straight-lining remains significant at the respondent level in six countries (Belgium, Ireland, the Netherlands, Slovenia, Switzerland, and the United Kingdom) out of the eight in which the correlation is significant at the interview level (without decomposition). This may be an indication that the relationship between interview speed and straight-lining is influenced by the respondent, the interviewer, and their interaction. Moreover, the correlation is positive and significant at the interviewer level in five countries (Estonia, Ireland, Slovenia, Switzerland, and the United Kingdom). In these countries, the interviewer behavior conjointly influences the interview speed and the straight-lining behavior. Given that speed and straight-lining are both indicators of satisficing, we may hypothesize that the interviewer behavior induces satisficing behavior from the respondent. Although we can observe that the relationship between straight-lining tendency and interview speed is not very strong overall, the results in table 6 provide empirical proof that this relationship does exist at the interviewer level in some countries. 7. DISCUSSION The aim of this paper was to validate previous results, which show large interviewer effects on interview speed and smaller but substantial interviewer effects on straight-lining tendency, at least in some countries. The analysis was then taken a step further by examining the correlation between interview speed and straight-lining tendency, which are both indicators of satisficing. This correlation was then separated into the interviewer level and the respondent level using a multilevel multivariate model. The analyses were performed using data from fifteen countries that participated in the ESS7 using CAPI. Our results confirm the findings in previous research (Olson and Peytchev 2007; Loosveldt and Beullens 2013a, 2013b, 2017), offering support for external validation. Interviewer effects on interview speed in the ESS7 are large, ranging from 0.08 to 0.37 across countries, even after controlling for respondent characteristics. It is notable that the variability in interview speed is explained to a major extent by differences between interviewers. Interview speed seems to be a powerful variable to detect deviations from standardized interviews and to analyze other characteristics of data quality. The interviewer effects on straight-lining tendency are considerably smaller, ranging from 0.00 to 0.21, but still substantial in some countries, showing again the possible impact of interviewers on a data quality indicator for which the respondent and response behavior are usually considered responsible. Turning to the relationship between interview speed and straight-lining tendency, the correlation between the two indicators of satisficing is low, but positive and significant in seven of the fifteen countries examined. This supports the hypothesis that high interview speed and straight-lining are related to each other, probably through both interviewer and respondent satisficing behavior. Separating the correlation into interviewer and respondent levels within countries, however, shows a significant correlation at the respondent level in nine countries. In these, fast respondents—who are possibly giving straight-lined answers—might influence the interviewer to increase the interview speed, because the respondent wants to finish the interview quickly. More importantly, the interviewer-level correlation is significant in five countries. This result implies that in these countries, interviewers who carry out interviews at a high speed are also the ones who are responsible for interviews with a higher straight-lining tendency and vice versa. Slovenia, Switzerland, and the United Kingdom belong to the countries with a significant correlation at the interviewer level. This is surprising given that the intra-interviewer correlation of the percentage straight-lined answers is lower than 0.05 in these countries. This means that the small differences between interviewers in straight-lining tendency among their respondents is strongly linked to the speed at which they conducted the interview. It would be therefore interesting to look at other quality measures, such as acquiescence or the use of “don’t know” answers, and their relationship with interview speed at the respondent and interviewer level to check whether the same results can be observed. This study has some limitations. First, given the non-experimental nature of the data, we are not able to draw any conclusions about the causality of the relationship between interview speed and straight-lining tendency. An experiment in which interviewers are randomly assigned to two groups could address this issue. One group (half of the interviewers chosen randomly) would be instructed and trained to read the questions at a higher speed and to leave only short breaks, whilst the other group would be instructed and trained to read the questions more slowly and to leave longer breaks. If more straight-lined answers were found among the respondents interviewed by the first group than amongst the respondents interviewed by the second, interview speed could be shown to cause straight-lining. The questionnaire would, however, need to contain many blocks of items. Further, straight-lining only reflects one aspect of data quality. Different indicators could be considered, such as consistency or item nonresponse. One possibility would be to build a satisficing indicator such as the one constructed by Kaminska, McCutcheon, and Billiet (2010), and to study its relationship to interview speed at the respondent and the interviewer level. Lastly, a measurement of interview speed that is more directly linked to the measurement of straight-lining tendency would be preferable. Timers measuring the start and the end of blocks of items would probably provide a stronger proof of the relation between speed and straight-lining. To conclude, our analysis offers empirical proof of deviations from standardized interviewing and of the effect that interviewers may have on the quality of data. This brings into question the interviewer’s role in face-to-face surveys. Interviewers should facilitate the completion of the questionnaire and motivate respondents to optimize the response process and consequently give the best answers. However, interviewers rushing through the process can increase the difficulty of the task and reduce the data quality. Interview speed as a characteristic of the interviewer seems to be related to data quality indicators such as straight-lining and could, therefore, be used for fieldwork monitoring. Moreover, it might be necessary to integrate interviewers into analyses concerning response styles. Given that previous research (Olson and Peytchev 2007) shows that more-experienced interviewers perform faster interviews, the results imply that retraining these interviewers in standardized interviewing techniques may be necessary. The results also prove the importance of interviewer motivation, not only to gain the respondents’ participation but also to perform high quality interviews at the right speed rather than rushing through. Hence, interviewer training should also focus on interview speed, on top of focusing on doorstep interaction and keeping a neutral attitude. Further research should investigate in greater detail the relationship between interview speed and interaction characteristics. What aspects of interviewer behavior are responsible for the large differences between interviewers in interview speed? Which interaction characteristics have the largest effect on interview speed and/or on straight-lining tendency? The behavior of the interviewer can influence both the interview speed and the straight-lining tendency. Interviewer behavior characteristics such as whether the question is read exactly as written, the number of probes (Ongena and Dijkstra 2006), whether the interviewer apologetically pursues the reading of the remaining response options when the respondent interrupts with an answer before the end, whether the interviewer repeats the question when an inappropriate answer is given (Garbarski, Schaeffer, and Dykema 2016), or whether the interviewer gives clarification to the question (Schober, Conrad, Dijkstra, and Ongena 2012) can influence the interaction with the respondents and hence the interview speed. The respondent’s engagement and motivation (Garbarski et al. 2016), but also understanding of the question or discomfort with the answer, can also influence the interview speed. The use of “um” or laughter, as well as the fluency of speech or gaze aversion, can be signs of difficulty answering a question (Schober et al. 2012). To perform the required analysis, interview recordings would be needed or timers that differentiate between when the interviewer or the respondent is speaking. However, the best coding schemes to choose for the recordings and their level of precision would need to be defined (Ongena and Dijkstra 2006). Lastly, we believe that the correlation at the interviewer level between interview speed and straight-lining tendency discovered in this paper highlights the importance of interview speed as a quality indicator that is relatively easy to monitor during the fieldwork, but very little used. It seems worthwhile to investigate the relationship between interview speed and other data quality indicators. Supplementary Materials Supplementary materials are available online at http://www.oxfordjournals.org/our_journals/jssam/. Footnotes 1 http://www.europeansocialsurvey.org/data/download.html?r=7, last accessed October 2017. 2 We do not have information on the mother tongue itself, but we use the primary language spoken at home as a proxy. REFERENCES Bassili J. ( 1996 ), “The ‘How’ and ‘Why’ of Response Latency Measurement in Survey Research,” in Answering Questions: Methodology for Determining Cognitive and Communicative Processes in Survey Research , eds. Schwartz N. , Sudman S. , pp. 319 – 346, San Francisco : Jossey-Bass Publishers . Bassili J. N. , Scott B. S. ( 1996 ), “ Response Latency as a Signal to Question Problems in Survey Research ,” Public Opinion Quarterly , 60 , 390 – 399 . Google Scholar Crossref Search ADS Cannell C. F. , Miller P. V. , Oksenberg L. ( 1981 ), “ Research on Interviewing Techniques ,” Sociological Methodology , 12 , 389 – 437 . Google Scholar Crossref Search ADS Chang L. , Krosnick J. ( 2009 ), “ National Surveys via RDD Telephone Interviewing versus the Internet. Comparing Sample Representativeness and Response Quality ,” Public Opinion Quarterly , 73 , 641 – 678 . Google Scholar Crossref Search ADS Conrad F. G. , Tourangeau R. , Couper M. P. , Zhang C. ( 2011 ), “Interactive Interventions in Web Surveys Can Increase Response Accuracy,” paper presented at the 2011 Annual Meeting of the American Association for Public Opinion Research, Phoenix, AZ. Couper M. P. , Kreuter F. ( 2013 ). “ Using Paradata to Explore Item Level Response Times in Surveys ,” Journal of the Royal Statistical Society: Series A , 176 , 271 – 286 . Google Scholar Crossref Search ADS Crawford S. C. , Couper M. P. , Lamias M. J. ( 2001 ), “ Web Surveys ,” Social Science Computer Review , 19 , 146 – 162 . Google Scholar Crossref Search ADS ESS ( 2014 ), ESS Round 7: European Social Survey Round 7 Data (2014) Data file ( 1st ed. ), Norway : Norwegian Social Science Data Services, Data Archive and distributor of ESS data for ESS ERIC . ESS ( 2015 ), ESS Round 7: European Social Survey (2015) ESS-7 2014. Data Documentation Report (1st ed.) , Bergen : European Social Survey Data Archive, Norwegian Social Science Data Services for ESS ERIC Fowler F. J. , Mangione T. ( 1990 ), Standardized Survey Interviewing. Minimizing Interviewer‐Related Error , Newbury Park : SAGE Publications . Fricker S. , Galesic M. , Tourangeau R. , Yan T. ( 2005 ), “ An Experimental Comparisons of Web and Telephone Surveys ,” Public Opinion Quarterly , 69 , 370 – 392 . Google Scholar Crossref Search ADS Garbarski D. , Schaeffer N. C. , Dykema J. ( 2016 ), “ Interviewing Practices, Conversational Practices and Rapport: Responsiveness and Engagement in the Standardized Survey Interview ,” Sociological Methodology , 46 , 1 – 38 . Google Scholar Crossref Search ADS PubMed Grant J. T. , Mockabee S. T. , Monson J. Q. ( 2010 ), “ Campaign Effect on the Accessibility of Party Identification ,” Political Research Quarterly , 63 , 529 – 543 . Google Scholar Crossref Search ADS Groves R. M. ( 1989 ), Survey Error and Survey Costs , Hoboken, NJ : John Wiley & Sons . Heck R. H. , Thomas S. L. ( 2015 ), An Introduction to Multilevel Modeling Techniques: MLM and SEM Approaches using Mplus , ( 3rd ed.) , New York : Taylor and Francis . Heerwegh D. ( 2003 ), “ Explaining Response Latencies and Changing Answers Using Client-Side Paradata from a Web Survey ,” Social Science Computer Review , 21 , 360 – 373 . Google Scholar Crossref Search ADS Hox J. ( 1994 ), “ Hierarchical Regression Models for Interviewer and Respondent Effects ,” Sociological Methods and Research , 22 , 300 – 318 . Google Scholar Crossref Search ADS Hox J. , De Leeuw E. , Kreft I. ( 1991 ), “The Effect of Interviewer and Respondent Characteristics on the Quality of Survey Data: A Multilevel Model,” in Measurement Errors in Surveys , eds. Biemer P. P. , Groves R. M. , Lyberg L. E. , Mathiowetz N. A. , Sudman S. , pp. 439 – 461 , New York : John Wiley . Japec L. ( 2005 ), “Interviewer Burden and its Effects on Data Quality in the Swedish Part of the European Social Survey (ESS),” in Quality Issues in Interview Surveys, Some Contributions , ed. Japec L. , Stockholm : Department of Statistics Stockholm University . Johnson M. ( 2004 ), “ Timepieces: Components of Survey Question Response Latencies ,” Political Psychology , 25 , 679 – 702 . Google Scholar Crossref Search ADS Kaminska O. , McCutcheon A. L. , Billiet J. ( 2010 ), “ Satisficing among Reluctant Respondents in a Cross-National Context ,” Public Opinion Quarterly , 74 , 956 – 984 . Google Scholar Crossref Search ADS Krosnick J. A. ( 1991 ), “ Response Strategies for Coping with the Cognitive Demands of Attitude Measures in Surveys ,” Applied Cognitive Psychology , 5 , 213 – 236 . Google Scholar Crossref Search ADS Krosnick J. A. , Narayan S. , Smith W. R. ( 1996 ), “ Satisficing in Surveys: Initial Evidence ,” New Directions for Evaluation , 1996 , 29 – 44 . Google Scholar Crossref Search ADS Loosveldt G. , Beullens K. ( 2013a ), “ How Long Will It Take? An Analysis of Interview Length in the Fifth Round of the European Social Survey ,” Survey Research Methods , 7 , 69 – 78 . Loosveldt G. , Beullens K. ( 2013b ), “ The Impact of Respondents and Interviewers on Interview Speed in Face-to-Face Interviews ,” Social Science Research , 42 , 1422 – 1430 . Google Scholar Crossref Search ADS Loosveldt G. , Beullens K. ( 2017 ), “ Interviewer Effects on Non-Differentiation and Straightlining in the European Social Survey ,” Journal of Official Statistics, Special Issue on Total Survey Error (TSE) , 33 , 409 – 426 . Malhotra N. ( 2008 ), “ Completion Time and Response Order Effects in Web Surveys ,” Public Opinion Quarterly , 72 , 914 – 934 . Google Scholar Crossref Search ADS McClamroch K. J. ( 2011 ), “ Evaluating the Usability of Personal Digital Assistants to Collect Behavioral Data on Adolescent with Paradata ,” Field Methods , 23 , 219 – 242 . Google Scholar Crossref Search ADS Mulligan K. , Krosnick J. , Smith W. , Green M. , Bizer G. ( 2001 ), Non-Differentiation on Attitude Rating Scales: A Test of Survey Satisficing Theory, unpublished manuscript, Stanford University. Olson K. , Bilgen I. ( 2011 ), “ The Role of Interviewer Experience on Acquiescence ,” Public Opinion Quarterly , 75 , 99 – 114 . Google Scholar Crossref Search ADS Olson K. , Peytchev A. ( 2007 ), “ Effect of Interviewer Experience on Interview Speed and Interviewer Attitudes ,” Public Opinion Quarterly , 71 , 273 – 286 . Google Scholar Crossref Search ADS O’Muircheartaigh C. , Campanelli P. ( 1998 ), “ The Relative Impact of Interviewer Effects and Sample Design Effects on Survey Precision ,” Journal of the Royal Statistical Society: Series A , 1 , 63 – 77 . Google Scholar Crossref Search ADS Ongena Y. P. , Dijkstra W. ( 2006 ), “ Methods of Behavior Coding of Survey Interviews ,” Journal of Official Statistics , 22 , 419 – 451 . Pickery J. , Loosveldt G. ( 2004 ), “ A Simultaneous Analysis of Interviewer Effects on Various Data Quality Indicators with Identification of Exceptional Interviewers ,” Journal of Official Statistics , 20 , 77 – 89 . Revilla M. , Ochoa C. ( 2015 ), “ What Are the Links in a Web Survey among Response Time, Quality, and Auto-Evaluation of the Efforts Done ?” Social Science Computer Review , 33 , 97 – 114 . Google Scholar Crossref Search ADS Schnell R. , Kreuter F. ( 2005 ), “ Separating Interviewer and Sampling-Point Effects ,” Journal of Official Statistics , 21 , 389 – 410 . Schober M. F. , Conrad F. G. , Dijkstra W. , Ongena Y. P. ( 2012 ), “ Disfluencies and Gaze Aversion in Unreliable, Responses to Survey Questions ,” Journal of Official Statistics , 28 , 555 – 582 . Turner M. ( 2013 ), New Perspectives on Interviewer-Related Error in Surveys: Application of Survey Paradata , Southampton : University of Southampton, Faculty of Social and Human Sciences . Turner G. , Sturgis P. , Martin D. ( 2015 ), “ Can Response Latencies Be Used to Detect Survey Satisficing on Cognitively Demanding Questions ?” Journal of Survey Statistics and Methodology , 3 , 89 – 108 . Google Scholar Crossref Search ADS Vassallo R. , Durrant G. , Smith P. ( 2017 ), “ Separating Interviewer and Area Effects by using a Cross-Classified Multilevel Logistic Model: Simulation, Findings and Implications for Survey Designs ,” Journal of the Royal Statistical Society: Series A , 180 , 531 . Google Scholar Crossref Search ADS Wells T. , Rao K. , Link M. W. , Pierce C. ( 2012 ), “Flagging Speeders in a Multi-Mode (Mobile and Online) Survey,” paper presented at the 2012 Annual Meeting of the Midwest Chapter of the American Association for Public Opinion Research, Chicago, IL. West B. T. , Blom A. G. ( 2016 ), “ Explaining Interviewer Effects: A Research Synthesis ,” Journal of Survey Statistics and Methodology , 5, 175 – 211 . Yan T. , Tourangeau R. ( 2008 ), “ Fast Times and Easy Questions: the Effects of Age, Experience and Question Complexity on Web Survey Response Times ,” Applied Cognitive Psychology , 22 , 51 – 68 . Google Scholar Crossref Search ADS Zhang C. , Conrad F. G. ( 2014 ), “ Speeding in Web Surveys: The Tendency to Answer Very Fast and Its Association with Straightlining ,” Survey Research Methods , 8 , 127 – 135 . © The Author 2017. Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

Journal of Survey Statistics and MethodologyOxford University Press

Published: Dec 1, 2018

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off