Annex 2 Details of Regression Analysis and Output Tables
1 Analyses were run using SAS logistic procedure (backward selection) 25. Two main sets of logistic regression analysis were run on use of childcare providers; using either only the demographic factors or using the demographic factors and the reasons that parents had for using childcare. Only those factors which were significant in predicting childcare use are included in the tables.
2 Logistic regression is a multivariate statistical technique that uses a set of independent variables to predict the probability of an event occurring. The odds of having a particular outcome are modelled. In this report, the models estimate the odds that a household with particular characteristics used childcare. Odds are calculated as p/(1-p) where p is the proportion or percentage having the characteristic of interest. For example if 20% of a specific group received childcare, then the odds of receiving childcare for members of this group are: 0.2/0.8=0.25 (or 0.25:1). If within another group 40% received childcare then the odds for this group are 0.4/0.6=0.67 (or 0.67:1). We can then compare the groups by comparing the odds. The odds of receiving childcare for the second group are 2.68 (0.67/0.25) times higher than for the first group. This is the 'odds ratio'. Therefore to calculate the odds ratios each subgroup ( e.g. type of household) is compared with a 'control group'.
3 The control groups for the analysis of the childcare data were chosen as those with the highest number of respondents. For example, 'Highlands and Islands' was chosen for the control group for area because the combination of local authorities within this group led to it having a higher number of respondents than either Glasgow or Edinburgh.
4 The model quality was assessed using the R-Sqd (Max) which represents how much of the data variability (of the dependent variable) is explained by the model. In the current analysis, less than 20% was seen to be a poor quality model, 20-39% an average quality model, 40% - 59% a decent quality model, and 60% and over a good quality model. The interactions between variables were examined, but none of these improved the quality of the models (in terms of R-Sqd) by more than 2% and so these were not included, in order to keep the models as simple as possible. It should be noted that the models relating to parents' friends, relative/partner or registered childminder as childcare provider were poor and the results are therefore not presented in detail.
5 A 'poor' model is one where the variables available for analysis do not explain much of the variation in the dependant variable. For example, in the analyses presented in this report, the model for childcare by a friend was 'poor' suggesting that, although some of the demographic factors in the analysis were predictors of that type of childcare, they explained only a small part of the variation between households in the use of that type of childcare and therefore other factors not measured in the survey explained more. In contrast, the model for nursery use was 'good' indicating that the demographic factors which were significant in the model (age of child and using childcare for child's development) explained most of the variation between households in their use of this type of childcare.
6 The tables for each type of childcare include the odds ratio and p-value for each variable (except the control group). An odds-ratio of 1 means that there is no difference between the two groups, an odds ratio of more than 1 means that the group has greater odds of receiving (that type of) childcare than the control group, and an odds ratio of less than 1 means that the group has lower odds of receiving (that type of) childcare than the control group. The p-values indicate whether an odds ratio is significantly different from what would have been expected to be found by chance ( i.e. if there was no relationship between the variable and the outcome). A small p-value (less than 0.05) suggests that the true odds ratio is statistically different from 1.
7 In the tables in this Appendix the control groups are shown in italics. The odds ratios and p-values are given and a comment added to indicate whether each subgroup is significantly different from the control group. For example in Table 1 for the characteristics 'household type', small family was chosen as the control group because there were more families of this type than of the other types. The results show that large families have lower odds of using childcare than small families (0.70:1) and that single parent families have higher odds of using childcare than small families (1.49:1). As the p-values for both of these are less than 0.0001, we can say that these differences are statistically significant. Therefore it is concluded that household type is a predictor of using childcare. In contrast, in the same table it can be seen that compared with the control group of 'White Scottish', the odds of using childcare are 0.82:1 for 'White-other background; and 0.73:1 for 'Any other background' but as the p-values are large (greater than 0.05) these differences are not statistically significant. Therefore it is concluded that ethnicity of not a predictor of childcare use in this model.
Tables
Table 1: Predictors of children receiving any type of childcare, R-Square = 0.297
Table 2: Demographic characteristics of children receiving and not receiving childcare
Table 3: Predictors of children receiving informal childcare only, R-Square = 0.403
Table 4: Predictors of children receiving formal childcare only, R-Square = 0.276
Table 5: Demographic characteristics of children receiving formal and informal childcare
Table 6: Predictors that children would attend nursery or playgroups, R-Square = 0.666
Table 7: Demographic characteristics of children who attend and do not attend nursery/playgroups
Table 8: Predictors of children receiving Out of School Care ( OSC), R-Square = 0.340
Table 9: Demographic characteristics of children who attend and do not attend Out of School Care
Table 10: Demographic characteristics of children cared for and not cared for by registered childminders
Table 11: Demographic characteristics of children receiving and not receiving childcare from parent's friends
Table 12: Demographic characteristics of children cared for and not cared for by relative/partner of HiH
Table 13: Reasons for using childcare by demographic characteristics
Table 14: Predictors of parents being satisfied with quality of childcare, R-Square = 0.477