Exploratory factor analyses of each of the 11 original scales separately indicated there were 29 subscales with 2 to 13 items per subscale; three subscales had 10 or more items; 12 subscales had 4 items or less. Model fit was acceptable in most cases. Cronbach’s alphas for the subscales ranged from 0.13 to 0.92 with 17 being 0.70 or higher. Most alphas <0.70 included only three or four items, but acceptable average inter-item correlations . Twenty-five of 29 subscales significantly bivariately correlated with composite effective or ineffective VPP.
To our knowledge, this is the first report of the psychometric characteristics of theory based scales and subscales to predict a parent’s use of VPP. Most studies using TPB  or MGDB [12–15] used single dimensional scales for each predictive construct. Our approach, alternatively, found single dimensions did not adequately fit the items for most scales/constructs. Using the scree plot criterion and interpretability, exploratory analyses obtained one to four dimensions per scale/construct.
A number of subscales (12/29) had internal reliabilities less than 0.7 which is generally considered low . Low scale reliability attenuates relationships with other variables . Most of these subscales included only 3 or 4 items. Since Cronbach’s alpha is sensitive to the number of items, for subscales with few items an average inter-item correlation in the range of 0.15 to 0.50 is considered an indicator of an acceptable level of internal consistency . Of the 12 subscales with 4 items or less, the average inter-item correlation was in the acceptable range for 9 of them, and for 2 it exceeded the range. This suggests that a true dimension was detected, but additional work is needed to generate new items to expand the subscale, test dimensionality, and re-assess the psychometrics of the new subscales and scales. Since norms have a long history as a part of the Theory of Planned Behavior , the Descriptive Norms subscale should be retained, but further developed to enhance its reliability.
Factorial validity (CFA) could not be established for four scales even though internal consistency reliability was acceptable for all but the Autonomy scale. The CFA for the Autonomy items could not achieve positive definite status. Several direct estimation methods (weighted least squares, mean-adjusted weighted least squares, and variance-adjusted weighted least squares) were tried, but to no avail. The low Cronbach’s alpha (0.31), the consistently low corrected item total correlations (0.15, 0.19, 0.25), and the low average inter-item correlation (0.17) suggested that autonomy is a complex construct and the items we included tapped multiple dimensions, which were not highly interrelated. Since Autonomy included only three items, more development of this scale and possible subscales is warranted.
We had no theoretical foundation for theoretically deducing which MGDVPP subscales would correlate with EVPP or IVPP. Despite some low reliabilities, 25 of 29 subscales correlated with one or the other of the composite EVPP or IVPP. Parent Values (a Relatedness subscale) significantly inversely correlated with EVPP and IVPP. Similarly, most Intentions subscales inversely correlated with EVPP and IVPP. It is likely that respondents did not know which VPP were effective or ineffective, which may have influenced these relationships. It is possible that respondents thought the Intention items should only be answered positively if they were not already doing it, but intending to do it in the next month. Future research with these scales will need to address these issues.
Thirty intercorrelations among subscales were tested; 9 were not significant; 5 were significant at p < 0.05, 1 at p < 0.01, and 15 at p < 0.001. The subscales tended to be intercorrelated in expected directions within scales. The highest correlation was 0.51 between the Perceived Barriers of Respondent Doesn’t Like Vegetables and Cost of Vegetables. Intersubscale correlations will need to be validated in future studies. While not high enough in this sample to constitute multicollinearity, it is possible that future studies will identify different dimensions combining subscales in the current sample.
The strengths of this research include use of a broad innovative theoretical model to predict behaviors (here vegetable parenting practices); qualitative methods to generate items from the target group; and narrowly focused on parents of a developmentally similar age group. A number of limitations exist. The sample was limited in size and diversity. Further research is needed with larger samples to permit more sophisticated analyses and with more diverse samples to test generalizability across gender, ethnicity, and socioeconomic status. The internet survey method did not allow collecting and matching data from a second time point, thereby precluding an assessment of test-retest reliability; and the same sample was employed for exploratory and confirmatory factor analyses. Predictive validity was tested with cross-sectional data; these need to be verified with longitudinal data. Additional research with larger samples should use Item Response Modeling (IRM) to better understand the sequencing of items, difficulties across the latent constructs, the matching of item distributions with participant distributions, and to assess differences in item responses (i.e. differential item functioning) by demographic characteristics [30, 31]. IRM would also permit efficient reduction of items in the subscales with larger numbers by identifying items redundant at location along the latent variable . Twenty-nine subscales were identified. While model testing research should include all 29 to verify (or disconfirm) the current findings, investigators with a more practical or applied intent may wish to select subscales most clearly related to their efforts. The four subscales that did not correlate with EVPP or IVPP, and the ones that correlated in unexpected directions, need further testing in other samples.
Although further development is warranted, these scales and subscales can be used in studies attempting to understand why parents might use effective and ineffective vegetable parenting practices.