beta

You're viewing our new website - find out more

Publication - Research publication

The microsegmentation of the autism spectrum: research project

Published: 26 Mar 2018

Economic research on autism and implications for Scotland, including how the economic cost of autism can inform strategy and planning.

357 page PDF

4.7 MB

357 page PDF

4.7 MB

Contents
The microsegmentation of the autism spectrum: research project
Footnotes

357 page PDF

4.7 MB

Footnotes

1 We have used the term ‘co-occurring conditions’ rather than ‘comorbidities’ throughout the report. Feedback from autistic people has indicated their view that it is overly medical, and some of the issues referred to are not of a medical nature. Both terms are currently used in the world literature.

2 An event per variable ( EPV) rate describes the relationship between the number of variables included in a logistic regression model and the smallest number of dependent variable event outcomes (i.e. the number of events associated with the least frequent binary category [ LFBC]). All models reported in Chapter 7 adhere to an EPV rate of 10, indicating that for every 10 events/outcomes associated with the LFBC, an additional independent variable could be included in the model (e.g. 20 events associated with the LFBC would allow two independent variables to be included in the model, and 50 events associated with the LFBC would allow for the inclusion of five independent variables in the model. Ensuring that all models meet an EPV ratio of 10 reduces the likelihood of type 1 errors.

3 Only 16 individuals were recorded as having Tourette’s Syndrome. It was decided that the categories of OCD and Tourette’s should be combined for the purposes of analysis. A close relationship has long been recognised between the two conditions, not only in functional but also in terms of possible aetiological correlates (Liu et al., 2015; Lombroso & Scahill, 2008; Mell, Davis, & Owens, 2005; Pauls et al., 1986).

4 A value of .02 was utilised here as a cut-off for Nagelkereke R 2 change as predictors associated with this level of improvement in the model were found to be both statistically significant and also associated with a Wald statistic sufficiently large to indicate that the predictor was making a significant contribution to the model.

5 There was no evidence that any of the variables included in the final model were collinear with the standard errors of each predictor less than 1, and changes in the b coefficients associated with each predictor less than .1 with the addition of each new predictor. However, there were 5 cases in which standardised residuals were > 2 or Cook’s distances were < 1, therefore these cases were removed and the analysis was re-run. This adjusted analysis resulted in an improvement of the classification accuracy of the model of > 2% therefore it is the results of this adjusted model which have been reported above, the original model has been included in Appendix C.5.

6 A value of .02 was utilised here as a cut-off for Nagelkereke R 2 change as predictors associated with this level of improvement in the model were found to be both statistically significant and also associated with a Wald statistic sufficiently large to indicate that the predictor was making a significant contribution to the model.

7 There was no evidence that any of the variables included in the final model were collinear, with the standard errors of each predictor less than 1, and changes in the b coefficients associated with each predictor less than .2 with the addition of each new predictor. However, 8 responses were associated with Cook’s values which exceeded 1 and studentised residuals which exceeded 2. When these cases were removed from the analysis this led to a > 1% improvement of the amount of variance explained by the model, therefore the above table reports the results of the original analysis including all cases. The original model including all cases has been included in Appendix C.5.

8 This total reflects the total number of individuals for whom information on transitions between preschool and primary school were available

9 This total reflects the total number of individuals for whom information on transitions between primary school and secondary school were available

10 There was no evidence that any of the variables included in the final model were collinear with the standard errors of each predictor less than 1, and changes in the b coefficients associated with each predictor less than .2 with the addition of each new predictor. However, 6 responses were associated with Cook’s values which exceeded 1 and studentised residuals which exceeded 2. Removing these cases from the analysis resulted in an improvement of > 2% in the classification accuracy of the model and therefore it is the results of this adjusted analysis which has been reported above. The original model including all cases has been included in 7.2.4

11 There was no evidence that any of the variables included in the final model were collinear with the standard errors of each predictor less than 1, and changes in the b coefficients associated with each predictor less than .2 with the addition of each new predictor. However, 15 responses were associated with Cook’s values which exceeded 1 and studentised residuals which exceeded 2. Removing these cases from the analysis resulted in an improvement of > 2% in the classification accuracy of the model and therefore it is the results of this adjusted analysis which has been reported above. The original model including all cases has been included in 7.2.6

12 Note: similar but positive results were also found in analysis focussing on the presence of any mood disorder diagnosis rather than just depression, X 2 (1, 384) = 36.57, p < .001, and follow-up odds ratio statistics indicated that those living independently were 3.88 times more likely to have a mood disorder.

13 There was no evidence that any of the variables included in the final model were collinear with the standard errors of each predictor less than 1, and changes in the b coefficients associated with each predictor less than .2 with the addition of each new predictor. However, 8 responses were associated with Cook’s values which exceeded 1 and studentised residuals which exceeded 2. Removing these cases from the analysis resulted in an improvement of > 2% in the classification accuracy of the model and therefore it is the results of this adjusted analysis which has been reported above. The original model including all cases has been included in 7.2.7

14 There was no evidence that any of the variables included in the final model were collinear with the standard errors of each predictor less than 1, and changes in the b coefficients associated with each predictor less than .2 with the addition of each new predictor. Residual checks were carried out for this model, however all Cook’s distances were found to be < 1 and all studentised residuals were found to be < 2


Contact