Abstract Objective The Self-Efficacy for Diabetes Scale (SED) is a widely used measure of diabetes-specific self-efficacy with three subscales: diabetes-specific self-efficacy (SED-D), medical self-efficacy (SED-M), and general self-efficacy (SED-G). The present study examined the factor structure and construct validity of the SED in 116 youth, aged 10–16 years (13.60 ± 1.87), with type 1 diabetes mellitus. Methods Confirmatory factor analysis (CFA) was used to examine the factor structure of the SED. Correlational and regression analyses examined relations between subscales and select outcomes. Results CFA of the original three-factor structure provided a poor fit to the data. Factor models using rescaled items were tested. Results provided preliminary evidence for the SED-D as an independent one-factor model, and for a reduced one-factor model. Significant associations were found between the SED subscales, responsibility for diabetes management, and glycated hemoglobin. Conclusions Results provide limited support for the SED-D as a reliable and valid measure of diabetes-specific self-efficacy. assessment, confirmatory factor analysis, pediatric, self-efficacy, type 1 diabetes mellitus The management of type 1 diabetes mellitus (T1DM) requires considerable knowledge and skill, and puts a significant strain on individuals and families. Although effective management of T1DM is highly correlated with the prevention of diabetes complications (Hood, Peterson, Rohan, & Drotar, 2009), many youth with T1DM have glycated hemoglobin (HbA1c) levels above the target (Beck et al., 2012). Many individual factors have been examined as predictors of T1DM management in youth, with numerous studies supporting the positive association between self-efficacy and diabetes self-care (Battaglia, Alemzadeh, Katte, Hall, & Perlmuter 2006; Grey et al., 2009; Kaugars, Kichler, & Alemzadeh, 2011; Whittemore, Jaser, Chao, Jang, & Grey, 2012). Consistent with Bandura’s conceptualization, self-efficacy for diabetes intends to measure one’s belief in their ability to engage in health-promoting behaviors consistent with T1DM management. Measures of diabetes self-efficacy in youth have been shown to be associated with diabetes self-care (e.g., youth and parent self-report), frequency of blood glucose checks, and HbA1c in cross-sectional studies (Battaglia et al., 2006; Grey et al., 2009; Kaugars et al., 2011; Whittemore et al., 2012), and with frequency of blood glucose checks and HbA1c in longitudinal research (Herge et al., 2012). Although these results suggest predictive value to measuring self-efficacy, many self-efficacy measures have been used in child and adolescent T1DM research, limiting the generalizability of findings and the ability to compare results across studies. For example, a recent review (Rasbach, Jenkins, & Laffel, 2015) identified ten different self-efficacy instruments that have been used in T1DM research among youth. By far the most commonly used measure of pediatric diabetes self-efficacy is the Self-Efficacy for Diabetes Scale (SED; Grossman, Brink, & Hauser, 1987). In fact, since 2010, the SED has been cited >60 times in the medical literature and it has been adapted for use in other populations and settings (Armstrong et al., 2011; Grey et al., 2013; Jaser et al., 2014; Kaugars et al., 2011; Kichler, Kaugars, Ellis, & Alemzadeh, 2010; Neylon et al., 2016; Rasbach et al., 2015). However, despite its popularity, there are some challenges to using the SED in its current form. First, no study to date has reported on the factor structure of this instrument using confirmatory factor analysis (CFA). This is potentially problematic because the SED proposes to have three underlying subscales hypothesized as integral components of diabetes self-efficacy, which include self-efficacy specific to diabetes, self-efficacy specific to medical situations, and self-efficacy in general situations. Without a CFA, it is impossible to determine whether all three subscales are necessary or whether their respective items are reliable indicators of each subscale. Moreover, a detailed examination of the SED’s subscale factor structures is also important because the diabetes-specific subscale of the SED has frequently been used alone in research (Grey et al., 2009; Kaugars et al., 2011; Whittemore et al., 2012). Second, only limited reliability and validity data exist for the SED, as well as limited data regarding the relations between the three latent factor subscales, or with diabetes outcomes (e.g., HbA1c, self-monitored blood glucose [SMBG], and diabetes responsibility). Thus, the primary purpose of this study was to conduct a CFA of the original SED measure in a sample of a similar age range, including each of the three subscales described in its original publication (Grossman et al., 1987), and to examine latent associations between the various subscales of the SED and measures of diabetes outcomes (i.e., HbA1c, SMBG, and diabetes responsibility). In addition, in instances where the originally proposed factor structure did not provide a good fit to the data, we conducted post hoc exploratory analyses to examine alternative indicator groupings. Method Participants A total of 125 parent–youth dyads composed of one youth with T1DM and his/her custodial parent were recruited to participate. Youth were eligible if they were between 10 and 16 years old, had been diagnosed with T1DM for at least 6 months, and were English-speaking. Youth with a diagnosis of developmental delay (i.e., autism, cerebral palsy, or mental retardation), as well as youth experiencing any hospitalization within the past year for a psychological disorder, were excluded from this study. Of the 125 dyads recruited, 116 dyads completed study measures. Youth had a mean age of 13.60 ± 1.87, with the majority identifying as male (N = 63; 54.3%) and Caucasian (N = 102; 91.9%). Youth had been diagnosed with diabetes for an average of 5.48 years (SD = 3.49) and had a mean HbA1c value of 9.14 (SD = 2.18). Procedure Institutional review board approval was obtained from the participating institutions before enrolling participants. Participants were recruited during routine diabetes clinic visits. They were asked to complete the web-based study measures (primarily using iPads during their clinic visit) and were compensated $25 for their participation. Youth’s HbA1c level and SMBG were obtained during the clinic visit and gathered for the study via medical chart review. The data used in this study are part of a larger study examining parent and child variables in T1DM management. Measures Self-Efficacy for Diabetes Scale The SED is a 35-item scale designed to measure youth’s perceived confidence and ability to manage diabetes, originally piloted with 68 children who were 12–16 years old (Grossman et al., 1987). This measure presupposes three subscales: self-efficacy specific to diabetes (SED-D; 24 items), self-efficacy specific to medical situations (SED-M; 5 items), and self-efficacy in general situations (SED-G; 6 items). The items ask youth to report how much they believe they can or cannot do things such as “avoid having low blood sugar reactions,” and each item is rated on a 6-point scale ranging from 1 (very sure I can’t) to 6 (very sure I can). In the original publication (Grossman et al., 1987), the authors did not identify which items were assigned to each subscale but we were able to obtain a copy of the original scoring procedures from another author (R. Whittemore, personal communication, June 25, 2015), which we used for our analyses. For the present study, all 35 original items were included. For the SED-D, Items 1–10, 12–16, 18, 20, 23, 24, 26, 31–33, and 35 were included. For the SED-M, items 11, 17, 19, 22, and 34 were included. For the SED-G, items 21, 25, and 27–30 were included. Diabetes Family Responsibility Questionnaire Youth completed the 17-item Diabetes Family Responsibility Questionnaire (DFRQ) to assess the distribution of diabetes-related responsibilities among a child and their caregivers (Anderson, Auslander, Jung, Miller, & Santiago, 1990). For each item, participants indicate whether the diabetes management task is primarily the responsibility of the child, caregiver, or is shared. Higher scores indicate increasing levels of child responsibility. Previously, the DFRQ has demonstrated adequate internal consistency (Lewin et al., 2006); the internal consistency in the present study was 0.84. Statistical Analyses CFAs were conducted in Mplus version 7.31 (Muthén & Muthén, 1998-2012). Monte Carlo power analyses indicated that the sample size of 116 was sufficient to test study aims (Muthén & Muthén, 2002). Given the items were six-level Likert-like items, all indicators were modeled as categorical using weighted least-squares mean and variance-adjusted (WLSMV) estimation. WLSMV estimation uses pairwise deletion in cases of missing data; fortunately, only 0.6% of the data in this study were missing. Multiple model fit indices were used for each CFA: chi-square, the comparative fit index (CFI), the Tucker–Lewis index (TLI), the root mean square error of approximation (RMSEA), and standard benchmarks for fit were applied (Browne & Cudeck, 1993; Hu & Bentler, 1999; Kline, 2005). Mplus version 7.31 (Muthén & Muthén, 1998–2012) was also used to conduct reliability and validity analyses of the SED (Holmbeck et al., 2008). We assessed the internal consistency of the SED using alpha coefficients. We assessed multicollinearity by using the variance inflation factor (VIF) and tolerance using SPSS Version 22 (IBM Corporation, 2013), with VIF values > 10 and tolerance values < 0.10 as our benchmarks (Hair et al., 1998). Convergent- and criterion-related validity of the SED was examined through latent correlations among the SED components and the DFRQ, and between the SED components and diabetes outcomes (i.e., youth’s HbA1c and SMBG), respectively. Additional evidence for criterion-related validity was examined using a regression analysis to predict HbA1c. In the first step of the regression, youth’s age, gender, and duration of diagnosis were entered; in the second step, the three SED subscales were entered. We hypothesized that the SED subscales would be significantly associated with the DFRQ, HbA1c, and SMBG. Results Distribution of Items A priori, the distribution of the original 35 items of the SED was examined and a negative skew was identified for multiple items. Specifically, <5% of the sample endorsed response categories 1–3 (ranging from Very Sure I Can’t to Somewhat Sure I Can’t) for 37.5% of SED-D items, 40% of SED-M items, and 33% of SED-G items. Because this skew indicates a potential problem with response variability, we opted to rescale items such that low (i.e., <5% of the total sample) endorsement response categories were collapsed, which improves fit indices when modeling variables with significant skew at the latent level (Agresti, 2010; Reise et al., 2011). This approach resulted in 28 rescaled items across the three scales, with the resulting items having a minimum of three response categories and a maximum of five response categories. Using these rescaled items, we proceeded with the CFA. Confirmatory Factor Analyses Using an iterative approach, the CFA is examined: (a) three one-factor models (i.e., an independent model for each of the three SED constructs) and (b) a three-factor model (consistent with the original SED). Additionally, as described below, some items did not significantly load on the hypothesized latent variable, and thus, alternative factor structures that dropped problematic items were also estimated (see Table I and description below, and Table III for a list of study items). Notably, in the course of conducting our initial CFAs, there was some indication that two additional models (a one-factor model and a two-factor model) might provide an adequate fit, and therefore, we examined these as well. Table I. Model Fit Indices for Raw and Rescaled Items One-factor SED-D Reduced one-factor SED-D One-factor SED-M One-factor SED-G Reduced one-factor SED-G Three-factor Reduced two-factor Reduced one-factor Itemsa 24 22 5 6 5 35 33 33 ÷2(df) 322.94/252 272.19/209 24.69/5 8.71/9 6.76/5 713.67/557 629.01/494 631.95/495 P <0.01 <0.01 <0.01 0.46 0.24 <0.01 <0.01 <0.01 CFI 0.92 0.93 0.83 1.00 0.97 0.89 0.90 0.90 TLI 0.92 0.93 0.65 1.00 0.94 0.88 0.90 0.90 RMSEA 0.05 (0.03–0.06) 0.05 (0.03–0.07) 0.18 (0.12–0.26) 0.00 (0.00–0.10) 0.06 (0.00–0.15) 0.05 (0.04–0.06) 0.05 (0.04–0.06) 0.05 (0.04–0.06) WRMR 0.90 0.88 0.82 0.43 0.42 1.01 0.98 0.99 One-factor SED-D Reduced one-factor SED-D One-factor SED-M One-factor SED-G Reduced one-factor SED-G Three-factor Reduced two-factor Reduced one-factor Itemsa 24 22 5 6 5 35 33 33 ÷2(df) 322.94/252 272.19/209 24.69/5 8.71/9 6.76/5 713.67/557 629.01/494 631.95/495 P <0.01 <0.01 <0.01 0.46 0.24 <0.01 <0.01 <0.01 CFI 0.92 0.93 0.83 1.00 0.97 0.89 0.90 0.90 TLI 0.92 0.93 0.65 1.00 0.94 0.88 0.90 0.90 RMSEA 0.05 (0.03–0.06) 0.05 (0.03–0.07) 0.18 (0.12–0.26) 0.00 (0.00–0.10) 0.06 (0.00–0.15) 0.05 (0.04–0.06) 0.05 (0.04–0.06) 0.05 (0.04–0.06) WRMR 0.90 0.88 0.82 0.43 0.42 1.01 0.98 0.99 Note. One-factor SED-D, diabetes-specific component modeled independently; Reduced one-factor SED-D, diabetes-specific component modeled independently removing Items #20 and #32; one-factor SED-M, medical component modeled independently; One-factor SED-G, general component modeled independently; Reduced one-factor SED-G, general component modeled independently removing Item #25; Three-factor, original three-component model, with each SED component modeled as a unique factor; Reduced two-factor, SED-M and SED-G combined into one factor; Reduced one-factor, all SED items modeled as a single factor. CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; WRMR = weight root mean square residual. a Items = the number of items included in the scale. Table III. Standardized Indicator Loadings and Error Variance SED-D 1. Be the one in charge of giving my insulin injection to myself 0.49 (0.09) 2. Figure out my own meals and snacks at home 0.63 (0.08) 3. Figure out what foods to eat when I am away from home 0.60 (0.08) 4. Keep track of my own blood sugar levels 0.80 (0.06) 5. Watch my own sugar levels in my urine 0.86 (0.05) 6. Change the amount of time I get insulin when I get a lot of extra exercise 0.66 (0.07) 7. Judge the amount of food I should eat before activities 0.50 (0.08) 8. Figure out how much insulin to give myself when I am sick in bed 0.60 (0.06) 9. Prevent having reactions 0.45 (0.08) 10. Avoid or get rid of dents, swelling, or redness of my skin where I get my shot 0.45 (0.08) 12. Suggest to my parents changes in my insulin dose 0.52 (0.08) 13. Sleep away from home on a class trip or at a friend’s house where no one knows about my diabetes 0.52 (0.09) 14. Keep myself free of high blood sugar levels 0.51 (0.08) 15. Know how to make my urine tests look better or worse than they are 0.43 (0.07) 16. Avoid having acetones 0.62 (0.07) 18. Feel able to stop a reaction when I am having one 0.49 (0.07) 20. Tell a friend I have diabetes 0.29 (0.13) 23. Prevent blindness and other complications from my diabetes 0.74 (0.06) 24. Tell my boyfriend or girlfriend I am diabetic 0.45 (0.11) 26 Get as much attention from others when my diabetes is under control as when it isn’t 0.56 (0.08) 31. Regularly wear a medical alert tag or bracelet which says I have diabetes 0.23 (0.09) 32. Sneak food not on my diet without getting caught 0.09 (0.10) 33. Believe that I have the ability to have control over my diabetes 0.63 (0.07) 35. Run my life the same as I would if I didn’t have diabetes 0.37 (0.09) SED-M 11. Talk to my doctor myself and ask for the things I need 0.60 (0.07) 17. Change my doctor if I don’t like him/her 0.45 (0.09) 19. Ask for help I need from other people when I feel sick 0.52 (0.09) 22. Argue with my doctor if I felt he/she were not being fair 0.58 (0.07) 34. Follow my doctor’s orders for taking care of my diabetes 0.75 (0.07) SED-G 21. Play baseball or other sports that take a lot of energy 0.43 (0.10) 25. Do things I have been told not to when I really want to do them 0.07 (0.10) 27. Easily talk to a group of people at a party when I don’t know them 0.55 (0.08) 28. Make a teacher see my point of view 0.65 (0.07) 29. Show my anger to a friend when he/she has done something to upset me 0.36 (0.08) 30. Take responsibility for getting my homework and chores done 0.58 (0.10) SED-D 1. Be the one in charge of giving my insulin injection to myself 0.49 (0.09) 2. Figure out my own meals and snacks at home 0.63 (0.08) 3. Figure out what foods to eat when I am away from home 0.60 (0.08) 4. Keep track of my own blood sugar levels 0.80 (0.06) 5. Watch my own sugar levels in my urine 0.86 (0.05) 6. Change the amount of time I get insulin when I get a lot of extra exercise 0.66 (0.07) 7. Judge the amount of food I should eat before activities 0.50 (0.08) 8. Figure out how much insulin to give myself when I am sick in bed 0.60 (0.06) 9. Prevent having reactions 0.45 (0.08) 10. Avoid or get rid of dents, swelling, or redness of my skin where I get my shot 0.45 (0.08) 12. Suggest to my parents changes in my insulin dose 0.52 (0.08) 13. Sleep away from home on a class trip or at a friend’s house where no one knows about my diabetes 0.52 (0.09) 14. Keep myself free of high blood sugar levels 0.51 (0.08) 15. Know how to make my urine tests look better or worse than they are 0.43 (0.07) 16. Avoid having acetones 0.62 (0.07) 18. Feel able to stop a reaction when I am having one 0.49 (0.07) 20. Tell a friend I have diabetes 0.29 (0.13) 23. Prevent blindness and other complications from my diabetes 0.74 (0.06) 24. Tell my boyfriend or girlfriend I am diabetic 0.45 (0.11) 26 Get as much attention from others when my diabetes is under control as when it isn’t 0.56 (0.08) 31. Regularly wear a medical alert tag or bracelet which says I have diabetes 0.23 (0.09) 32. Sneak food not on my diet without getting caught 0.09 (0.10) 33. Believe that I have the ability to have control over my diabetes 0.63 (0.07) 35. Run my life the same as I would if I didn’t have diabetes 0.37 (0.09) SED-M 11. Talk to my doctor myself and ask for the things I need 0.60 (0.07) 17. Change my doctor if I don’t like him/her 0.45 (0.09) 19. Ask for help I need from other people when I feel sick 0.52 (0.09) 22. Argue with my doctor if I felt he/she were not being fair 0.58 (0.07) 34. Follow my doctor’s orders for taking care of my diabetes 0.75 (0.07) SED-G 21. Play baseball or other sports that take a lot of energy 0.43 (0.10) 25. Do things I have been told not to when I really want to do them 0.07 (0.10) 27. Easily talk to a group of people at a party when I don’t know them 0.55 (0.08) 28. Make a teacher see my point of view 0.65 (0.07) 29. Show my anger to a friend when he/she has done something to upset me 0.36 (0.08) 30. Take responsibility for getting my homework and chores done 0.58 (0.10) Note. Items for each subscale are listed below subscale heading. SED-D = self-efficacy for diabetes-specific situations; SED-M = self-efficacy for medical situations; SED-G = self-efficacy for general situations. Error variances are presented in parentheses to the right of each standardized loading. In the first iteration, constructing three one-factor models, there was good model fit for the SED-D and excellent fit for the SED-G, but poor model fit for the SED-M (Table I). Notably, Items 32 (standardized loading = 0.07, p = .52) and 20 (standardized loading = 0.26, p = .06) did not significantly load onto the SED-D latent factor, and Item 30 (standardized loading = 0.17, p = .23) did not load onto the SED-G latent factor. Eliminating these items resulted in similar model fit for the SED-D, while the model fit for the SED-G appeared slightly worse (though still exhibited good fit to the data). In the second iteration, we examined a single three-factor model. The model revealed an acceptable RMSEA, but CFI and TLI values were just below the cutoff of 0.90. Again, Item 32 did not load onto the SED-D factor and Item 25 did not load onto the SED-G latent factor. This model also showed extremely large correlations across the three constructs: SED-D with SED-G = 0.93, p < .01; SED-D with SED-M = 0.91, p < .01; SED-M with SED-G = 0.99, p < .01, indicating multicollinearity. The third iteration examined a two-factor model. To construct this simplified two-factor model, items for the SED-M and SED-G subscales were combined to create one factor (which seemed reasonable given the r = .99 correlation found in the three-factor model), with the SED-D as the second factor. The model showed adequate fit to the data but also had a large correlation between the SED-D factor and the remaining (SED-M + SED-G) factor, r = .92, p < .01, indicating multicollinearity. Finally, in the last iteration, we examined a one-factor model using 33 SED items. This new one-factor model exhibited similar fit when compared with the two-factor model, and changes in CFI (0.903 − 0.902 = .001) and TLI (0.897 − 0.895 = 0.002) did not exceed 0.01, indicating minimal change in fit. Of note, the RMSEA of the null models for each of the last three models tested (the single three-factor model, the two-factor model, and the one-factor model with 33 items) was <0.158, which limits the degree to which incremental fit indices (e.g., CFI) are informative of model fit (Kenny, 2015). Reliability and Validity Using the scoring procedures adapted from previous studies (Kaugars, Kichler, & Alemzadeh, 2011; Whittemore, Jaser, Chao, Jang, & Grey, 2012), the SED-D (Cronbach’s α = .82) demonstrated adequate internal consistency, whereas the internal consistency of the SED-M (Cronbach’s α = .61) and SED-G (Cronbach’s α = .41) were poor (Nunnally & Bernstein, 1994). Although SED subscales were highly correlated, VIF (all values ≤ 1.40) and tolerance (all values ≥ 0.60) values for all subscales were within the acceptable range, which provides some evidence that these subscales may be distinct. The SED demonstrated good convergent validity based on significant correlations between SED subscales and the DFRQ (Table II). Further, significant associations among HbA1c levels and the SED-D, SED-M, and SED-G provide evidence of criterion-related validity (Table II). None of the SED subscales were significantly associated with SMBG, though both the SED-D and SED-G approached significance (0.066 and 0.052, respectively). Results of the regression analysis found that age (β = .39, t = 4.17, p ≤ .001) and SED-D (β = −.25, t = −2.16, p = .03) were predictors of HbA1c in the final model, thus providing additional support for criterion-related validity. Notably, the inclusion of SED subscales in the model accounted for an additional 8.3% of the variance. Table II. Intercorrelations and Descriptive Statistics Measure 1 2 3 4 5 6 7 8 9 1. SED-D — 2. SED-M .94*** — 3. SED-G .99*** 1.04*** — 4. DFRQ-G .40*** .35** .31* — 5. DFRQ-R .38*** .34** .36*** .78*** — 6. DFRQ-S .41*** .31** .45*** .79*** .59*** — 7. HbA1c −.56*** −.20* −.22* .07 .08 −.02 — 8. SMBG .17 .18 .17 −.15 −.41* .04 −.47*** — 9. Age .05 .07 .04 .53*** .50*** .23* .35*** −.40*** — Measure 1 2 3 4 5 6 7 8 9 1. SED-D — 2. SED-M .94*** — 3. SED-G .99*** 1.04*** — 4. DFRQ-G .40*** .35** .31* — 5. DFRQ-R .38*** .34** .36*** .78*** — 6. DFRQ-S .41*** .31** .45*** .79*** .59*** — 7. HbA1c −.56*** −.20* −.22* .07 .08 −.02 — 8. SMBG .17 .18 .17 −.15 −.41* .04 −.47*** — 9. Age .05 .07 .04 .53*** .50*** .23* .35*** −.40*** — Note. SED-D = self-efficacy for diabetes-specific situations, reduced model; SED-M = self-efficacy for medical situations; SED-G = self-efficacy for general situations, reduced model; DFRQ-G = diabetes responsibility for general health maintenance; DFRQ-R = diabetes responsibility for regimen tasks; DFRQ-S = diabetes responsibility for social presentation; HbA1c = glycated hemoglobin; SMBG = self-monitored blood glucose. * p < .05; **p < .01; ***p < .001. Discussion The SED is the most widely used measure of diabetes-specific self-efficacy in youth, but up until now there were no data available establishing its measurement properties (Rasbach et al., 2015). In our current analyses, we encountered a problem because a majority of the SED items were negatively skewed, which could impact its factor structure. Thus, we rescaled SED items to normalize their response variability before proceeding with primary analyses. Using these rescaled items, our CFAs did not support the original three-factor solution proposed by Grossman et al. (1987; see Table III for SED items and loadings). Our findings did, however, provide some support for the SED-D and SED-G individual one-factor models, but not for the SED-M, which suggests that it may be best to omit the SED-M in the future. Further, some support was found for a one-factor model that includes 33 items (items 25 and 32 excluded, as described above), although this one-factor model may include items that are not clinically relevant to youth and would potentially increase participant burden. Our reliability analyses indicated adequate internal consistencies for the SED-D and SED-G and poor internal consistency for the SED-M. These results are similar to the reliability estimates provided by Grossman and colleagues’ original publication (Grossman et al., 1987). In our study, examination of the latent correlations among the SED subscales revealed significant overlap (e.g., SED-D with SED-G, r = .93; SED-D with SED-M, r = .94; SED-M and the SED-G, r = 1.01). These findings are similar to those of Grossman et al. (1987), which were high (r = .81–.88) when corrected for attenuation. Considerable overlap between the subscales of the SED suggests that these subscales may not be measuring different latent constructs, but these results require replication to generalize broadly. We found support for the convergent validity of the SED by associating all three subscales with higher levels of youth responsibility for diabetes-related tasks (Kichler et al., 2010; Ott, Greening, Palardy, Holderby, & DeBell, 2000). There was also support found for criterion-related validity evidenced by the significant associations between SED subscale scores and youth’s HbA1c (Grossman et al., 1987; Hughes, Berg, & Wiebe, 2012). It is notable that a majority of SED items (i.e., 80%) demonstrated limited distributional variance. Rescaling items for latent variable analyses improved model fit, which suggests that researchers may need to closely evaluate item skew for the SED if it is used in future research, particularly for structural modeling approaches. Although it is possible that the present sample of youth generally possessed high levels of self-efficacy, this response pattern has emerged in several other studies using the SED (Armstrong, Mackey, & Streisand, 2011; Ott et al., 2000; Whittemore et al., 2012) and previous research has found a similar skewed response pattern for other diabetes-specific self-efficacy measures (Kappen, van der Bijl, & Vaccaro-Olko, 2001), as well as for a measure of headache medication self-efficacy (Seng, Nicholson, & Holroyd, 2016) and measures of academic self-efficacy (Diseth, Meland, & Breidablik, 2014; Smith, Wakely, De Kruif, & Swartz 2003; Toland & Usher, 2015). Although findings are inconsistent (Paunonen & Hong, 2010), some studies suggest that youth may report primarily high levels of self-efficacy to individual items because of impression management or the inclusion of seemingly low-challenge items (Bandura, 2006). In general, measures of self-efficacy may be vulnerable to response bias. Another possible explanation may be that the SED suffers from method effects because of the format of its response scale. How response sets are formatted is a concern in measurement development because it can influence how accurately responders are able to map their perceptions on to the most appropriate response option (Tourangeau, Rips, & Rasinski, 2000). Although the original SED uses a six-point response scale and provides labels for each response level, two components that are typically considered good practice, there is evidence across studies that youth may have difficulty using the original response scale. For example, the present data show that for 80% of the items, youth recorded responses using only three to five of the available responses. Recent research examining the psychometric properties of several academic self-efficacy measures suggest that a 4-point scale may be more appropriate for youth (Smith et al., 2003; Toland & Usher, 2015). This recommendation is also in line with the evidence suggesting that youth’s working memory capacity is limited to three to five categories (Cowan, 2010). Thus, one improvement to the SED may be to reduce the number of response options to no more than four, as this may improve response variability. However, another improvement to the SED’s response scale may be to change the actual response labels. This recommendation is supported by recent research, which emphasizes the importance of considering the use of response scale labels within a developmental framework (Mellor & Moore, 2014). Specifically, findings suggest that the utility of response labels relies heavily on how meaningful they are to respondents and that poorly worded labels or labels lacking an obvious ranking may be less reliable (Mellor & Moore, 2014; Weijters, Cabooter, & Schillewaert, 2010). In the case of the SED, the response labels included the following: 1 = Very sure I CAN NOT do; 2 = Somewhat sure I CAN NOT do; 3 = Sure I CAN NOT do; 4 = Sure I CAN do; 5 = Somewhat sure I CAN do; 6 = Very sure I CAN do. Given how youth’s responded, it is not clear that they could discriminate between somewhat sure and sure, leading to the negative skew. In the future, updating the response labels to make them more meaningful and discriminating for the target audience (i.e., youth) may improve item variability and the SED’s potential sensitivity. Limitations and Future Directions The present study had a number of strengths including a prospective examination of the SED’s relation with HbA1c and another measure of diabetes self-care, and a detailed examination of the SED’s factor structure. However, there are also limitations to consider when interpreting the results of the present study. First, because the majority of youth self-identified as Caucasian, the generalizability of the findings may be limited to youth from other racial backgrounds. Yet, the sample has a racial/ethnic diversity similar to that of the United States population of youth with T1DM (Liese et al., 2006). Thus, future confirmatory research should aim to extend the present findings to youth of other racial-ethnic backgrounds and a broader age span. Second, although the age and duration of diagnosis of the present sample was comparable with the sample of the initial validation of the SED (Grossman et al., 1987), it is unknown if these findings generalize to youth of different ages or diabetes duration. Future research should examine whether the present findings extend to youth of different ages and with a shorter duration of diagnosis. Given that divergent validity was not assessed in the current study, future research should examine this important aspect of measure validation. In addition to addressing these limitations, future research may also consider examining the age-appropriateness of the SED’s response scale. Ideally, potential follow-up studies would pilot a revised version of the SED-D scale, which would involve updating some of the items to be consistent with contemporary diabetes management. Moreover, it would be useful to pilot a four-point response scale to potentially improve item-response distributions and to compare this revised SED-D measure with other measures of self-efficacy in diabetes. Because the SED was developed in 1987, some of its items are less relevant for modern diabetes management (e.g., “urine testing”). Similar to other diabetes-specific self-efficacy measures, the SED also does not include items relevant to some aspects of contemporary diabetes management (e.g., using a continuous glucose monitoring system, insulin pump). Thus, future studies in pediatric diabetes may benefit from including contemporary items in an updated version of a self-efficacy measure for this population. Conclusion Establishing the measurement properties of the SED is essential before its continued use in scientific research and potential use in clinical care. The results of the present study did not provide support for Grossman et al.’s (1987) original scale construction and three-factor model, but rescaled and edited versions of the SED-D and SED-G were a good fit to the data, in addition to a one-factor model using 33 rescaled items. With regard to measure validity, correlations consistently demonstrated that the SED-D was more strongly associated with diabetes-related outcomes than the SED-G or SED-M. Likewise, the SED-D was more predictive of HbA1c than the one-factor model. Based on the present results, researchers are encouraged to be thoughtful about their use of the SED in future studies, particularly if they are considering using the full scale. Although a one-factor solution of an edited version of the SED demonstrated a good fit to the data, this measure includes items that may not be relevant in clinical settings and it demonstrated weaker associations with diabetes outcomes than the SED-D subscale alone. Thus, it may be more clinically meaningful to use items from the SED-D subscale, consistent with the approach used in many previous studies (Grey, Davidson, Boland, & Tamborlane, 2001; Grey et al., 2013; Kaugars et al., 2011). Nonetheless, clinicians may also find some benefit to using the reduced one-factor version of the entire scale. This would allow them to informally assess aspects of self-efficacy that are unrelated to diabetes, but that may be clinically meaningful to a particular patient as part of an idiographic assessment or personalized treatment planning. Finally, researchers are encouraged to evaluate the distribution and skew of individual study items in future use of the SED, as considerable skew can negatively impact model fit in latent variable analyses. Rescaling items with significant skew can improve the fit of latent/structural models, and does not necessitate making any changes to the administration of the scale a priori. Another strategy would be for future research to focus on scale refinement and resolve any ongoing problems related to response bias in the SED. Such refinement has the potential to make it easier to conduct future analyses, may improve the clinical utility of individual items, and would allow researchers to update some items to make them more consistent with modern diabetes management. Funding This research was supported in part by a grant from the Diabetes Institute of the University of Kansas Medical Center (to S.R.P.), by a grant R01-DK100779 (to S.R.P.) from the National Institutes of Health/National Institute of Diabetes and Digestive and Kidney Diseases, by the Marion and Donald Routh Student Research Grant Award from Division 54 of the American Psychological Association (to J.V.A.), and by the Pioneer Classes Dissertation Research Award from the Clinical Child Psychology Program of the University of Kansas (to J.V.A.). Conflicts of interest: None declared. References Agresti A. ( 2010). Analysis of ordinal categorical data . Hoboken, NJ: John Wiley & Sons. Google Scholar CrossRef Search ADS Anderson B. J. , Auslander W. F., Jung K. C., Miller J. P., Santiago J. V. ( 1990). Assessing family sharing of diabetes responsibilities. Journal of Pediatric Psychology , 15, 477– 492. doi: 10.1093/jpepsy/15.4.477 Google Scholar CrossRef Search ADS PubMed Armstrong B. , Mackey E. R., Streisand R. ( 2011). Parenting behavior, child functioning, and health behaviors in preadolescents with type 1 diabetes. Journal of Pediatric Psychology , 36, 1052– 1061. doi: 10.1093/jpepsy/jsr039 Google Scholar CrossRef Search ADS PubMed Bandura A. ( 2006). Guide for constructing self-efficacy scales. In Pajares F., Urdan T. (Eds.), Self-efficacy beliefs of adolescents (pp. 307– 337). Greenwich, CT: Information Age Publishing. Battaglia M. , Alemzadeh R., Katte H., Hall P. L., Perlmuter L. C. ( 2006). Disordered eating and psychosocial factors in adolescent females with type 1 diabetes mellitus. Journal of Pediatric Psychology , 31, 552– 556. doi: 10.1093/jpepsy/jsj047 Google Scholar CrossRef Search ADS PubMed Beck J. , Lewis T., Harrison D., Sternlof S., Comp C., Copeland K. ( 2012). Use of the mastery of stress instrument in caregivers of children newly diagnosed with type 1 diabetes: Identifying a need for further intervention. The Diabetes Educator , 38, 280– 288. doi: 10.1177/0145721711436133 Google Scholar CrossRef Search ADS PubMed Browne M. W. , Cudeck R. ( 1993). Alternative ways of assessing model fit. In Bollen K. A., Long J. S. (Eds.), Testing structural equation models (pp. 136– 162). Beverly Hills, CA: Sage. Cowan N. ( 2010). The magical mystery four: How is working memory capacity limited, and why? Current Directions in Psychological Science , 19, 51– 57. doi: 10.1177/0963721409359277 Google Scholar CrossRef Search ADS PubMed Diseth Å. , Meland E., Breidablik H. J. ( 2014). Self-beliefs among students: Grade level and gender differences in self-esteem, self-efficacy and implicit theories of intelligence. Learning and Individual Differences , 35, 1– 8. doi: 10.1016/j.lindif.2014.06.003 Google Scholar CrossRef Search ADS Grey M. , Davidson M., Boland E., Tamborlane W. V. ( 2001). Clinical and psychosocial factors associated with achievement of treatment goals in adolescents with diabetes mellitus. Journal of Adolescent Health , 28, 377– 385. doi: 10.1016/S1054-139X(00)00211-1 Google Scholar CrossRef Search ADS PubMed Grey M. , Whittemore R., Jaser S., Ambrosino J., Lindemann E., Liberti L., Northrup V., Dziura J. ( 2009). Effects of coping skills training in school-age children with type 1 diabetes. Research in Nursing and Health , 32, 405– 418. doi: 10.1002/nur.20336 Google Scholar CrossRef Search ADS PubMed Grey M. D. , Whittemore R., Jeon S., Murphy K., Faulkner M. S., Delamater A. ( 2013). Internet psycho-education programs improve outcomes in youth with type 1 diabetes. Diabetes Care , 36, 2475– 2482. doi: 10.2337/dc12-2199 Google Scholar CrossRef Search ADS PubMed Grossman H. Y. , Brink S., Hauser S. T. ( 1987). Self-efficacy in adolescent girls and boys with insulin-dependent diabetes mellitus. Diabetes Care , 10, 324– 329. Google Scholar CrossRef Search ADS PubMed Hair J. F. Jr. , Anderson R. E., Tatham R. L., Black W. C. ( 1998). Multivariate data analysis . Upper Saddle River, NJ: Prentice Hall. Herge W. M. , Streisand R., Chen R., Holmes C., Kumar A., Mackey E. R. ( 2012). Family and youth factors associated with health beliefs and health outcomes in youth with type 1 diabetes. Journal of Pediatric Psychology , 37, 980– 989. doi: 10.1093/jpepsy/jss067 Google Scholar CrossRef Search ADS PubMed Holmbeck G. N. , Thill A. W., Bachanas P., Garber J., Miller K. B., Abad M., Bruno E. F., Carter J. S., David-Ferdon C., Jandasek B., Mennuti-Washburn J. E., O'Mahar K., Zukerman J. ( 2008). Evidence-based assessment in pediatric psychology: Measures of psychosocial adjustment and psychopathology. Journal of Pediatric Psychology , 33, 958– 980. doi: 10.1093/jpepsy/jsm059 Google Scholar CrossRef Search ADS PubMed Hood K. K. , Peterson C. M., Rohan J. M., Drotar D. ( 2009). Association between adherence and glycemic control in pediatric type 1 diabetes: A meta-analysis. Pediatrics , 124, e1171– e1179. doi: 10.1542/peds.2009-0207 Google Scholar CrossRef Search ADS PubMed Hu L.-t. , Bentler P. M. ( 1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal , 6, 1– 55. Google Scholar CrossRef Search ADS Hughes A. E. , Berg C. A., Wiebe D. J. ( 2012). Emotional processing and self-control in adolescents with type 1 diabetes. Journal of Pediatric Psychology , 37, 925– 934. doi: 10.1093/jpepsy/jss062 Google Scholar CrossRef Search ADS PubMed IBM Corporation. ( 2013). IBM SPSS statistics for windows, version 22.0 . Armonk, NY: IBM Corporation. Jaser S. S. , Whittemore R., Chao A., Jeon S., Faulkner M. S., Grey M. ( 2014). Mediators of 12-month outcomes of two internet interventions for youth with type 1 diabetes. Journal of Pediatric Psychology , 39 3, 306– 315. doi: 10.1093/jpepsy/jsto81 Google Scholar CrossRef Search ADS PubMed Kappen M. J. , van der Bijl J. J., Vaccaro-Olko M. J. ( 2001). Self-efficacy in children with diabetes mellitus: Testing of a measurement instrument. Scholarly Inquiry for Nursing Practice , 15, 209– 221. Google Scholar PubMed Kaugars A. S. , Kichler J. C., Alemzadeh R. ( 2011). Assessing readiness to change the balance of responsibility for managing type 1 diabetes mellitus: Adolescent, mother, and father perspectives. Pediatric Diabetes , 12, 547– 555. doi: 10.1111/j.1399-5448.2010.00737.x Google Scholar PubMed Kenny D. A. ( 2015). Measuring model fit. Retrieved from http://davidakenny.net/cm/fit.htm#null Kichler J. C. , Kaugars A. S., Ellis J., Alemzadeh R. ( 2010). Exploring self-management characteristics in youths with type 1 diabetes mellitus: Does membership in a glycemic control category matter? Pediatric Diabetes , 11, 536– 543. doi: 10.1111/j.1399-5448.2010.00638.x Google Scholar CrossRef Search ADS PubMed Kline R. B. ( 2005). Principles and practice of structural equation modeling ( 2nd ed.), New York, NY: Guilford. Lewin A. B. , Heidgerken A. D., Geffken G. R., Williams L. B., Storch E. A., Gelfand K. M., Silverstein J. H. ( 2006). The relation between family factors and metabolic control: The role of diabetes adherence. Journal of Pediatric Psychology , 31, 174– 183. doi: 10.1093/jpepsy/jsj004 Google Scholar CrossRef Search ADS PubMed Liese A. D. , D'agostino R. B.Jr., Hamman R. F., Kilgo P. D., Lawrence J. M., Liu L. L., Loots B., Linder B., Marcovina S., Rodriguez B., Standiford D., Williams D. E.; SEARCH for Diabetes in Youth Study Group. ( 2006). The burden of diabetes mellitus among US youth: Prevalence estimates from the SEARCH for Diabetes in Youth Study. Pediatrics , 118, 1510– 1518. doi: 10.1542/peds.2006-0690 Google Scholar CrossRef Search ADS PubMed Mellor D. , Moore K. A. ( 2014). The use of Likert scales with children. Journal of Pediatric Psychology , 39, 369– 379. doi: 10.1093/jpepsy/jst079 Google Scholar CrossRef Search ADS PubMed Muthén L. K. , Muthén B. O. ( 1998-2012). Mplus user's guide ( 7th ed.). Los Angeles, CA: Muthén & Muthén. Muthén L. K. , Muthén B. O. ( 2002). How to use a Monte Carlo study to decide on sample size and determine power. Structural Equation Modeling: A Multidisciplinary Journal , 9, 599– 620. doi: 10.1207/s15328007sem0904_8 Google Scholar CrossRef Search ADS Neylon O. M. , Skinner T. C., O'Connell M. A., Cameron F. J. ( 2016). A novel tool to predict youth who will show recommended usage of diabetes technologies. Pediatric Diabetes , 17, 174– 183. Google Scholar CrossRef Search ADS Nunnally J. C. , Bernstein I. R. ( 1994). Psychometric theory . New York, NY: McGraw-Hill. Ott J. , Greening L., Palardy N., Holderby A., DeBell W. K. ( 2000). Self-efficacy as a mediator variable for adolescents' adherence to treatment for insulin-dependent diabetes mellitus. Children's Health Care , 29, 47– 63. doi: 10.1207/S15326888CHC2901_4 Google Scholar CrossRef Search ADS Paunonen S. V. , Hong R. Y. ( 2010). Self-efficacy and the prediction of domain-specific cognitive abilities. Journal of Personality , 78, 339– 360. doi: 10.1111/j.1467-6494.2009.00618.x Google Scholar CrossRef Search ADS PubMed Rasbach L. , Jenkins C., Laffel L. ( 2015). An integrative review of self-efficacy measurement instruments in youth with type 1 diabetes. The Diabetes Educator , 41, 43– 58. doi: 10.1177/0145721714550254 Google Scholar CrossRef Search ADS PubMed Reise S. P. , Ventura J., Keefe R. S., Baade L. E., Gold J. M., Green M. F., Kern R. S., Mesholam-Gately R., Nuechterlein K. H., Seidman L. J., Bilder R. ( 2011). Bifactor and item response theory analyses of interviewer report scales of cognitive impairment in schizophrenia. Psychological Assessment , 23, 245– 261. doi:10.1037/a0021501 Google Scholar CrossRef Search ADS PubMed Seng E. K. , Nicholson R. A., Holroyd K. A. ( 2016). Development of a measure of self-efficacy for acute headache medication adherence. Journal of Behavioral Medicine , 39, 1033– 1042. Advance online publication. doi: 10.1007/s10865-015-9683-9 Google Scholar CrossRef Search ADS PubMed Smith E. V. Jr. , Wakely M. B., De Kruif R. E. L., Swartz C. W. ( 2003). Optimizing rating scales for self-efficacy (and other) research. Educational and Psychological Measurement , 63, 369– 391. doi: 10.1177/0013164403063003002 Google Scholar CrossRef Search ADS Toland M. D. , Usher E. L. ( 2015). Assessing mathematics self-efficacy: How many categories do we really need? Journal of Early Adolescence , 36 7, 932– 960. doi: 10.1177/0272431615588952 Google Scholar CrossRef Search ADS Tourangeau R. , Rips L. J., Rasinski K. ( 2000). The psychology of survey response . Cambridge, UK: Cambridge University Press. Google Scholar CrossRef Search ADS Weijters B. , Cabooter E., Schillewaert N. ( 2010). The effect of rating scale format on response styles: The number of response categories and response category labels. International Journal of Research in Marketing , 27, 236– 247. doi: 10.1016/j.ijresmar.2010.02.004 Google Scholar CrossRef Search ADS Whittemore R. , Jaser S., Chao A., Jang M., Grey M. ( 2012). Psychological experience of parents of children with type 1 diabetes: A systematic mixed-studies review. Diabetes Educator , 38, 562– 579. doi: 10.1177/0145721712445216 Google Scholar CrossRef Search ADS PubMed © The Author 2017. Published by Oxford University Press on behalf of the Society of Pediatric Psychology. All rights reserved. For permissions, please e-mail: firstname.lastname@example.org
Journal of Pediatric Psychology – Oxford University Press
Published: Mar 1, 2018
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
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
All the latest content is available, no embargo periods.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud