Average treatment effect on treated (95% CI) based on inverse probability weighting using propensity score | Average treatment effect/adjusted risk difference (95% CI) based on logistic regression analysis | |||||
Empty (unadjusted) model* | Model adjusted for child and family variables† | Model adjusted for child and family variables and other parental strategies‡ | Empty (unadjusted) model § | Model adjusted for child and family variables ¶ | Model adjusted for child and family variables and other parental strategies** | |
Lifting | 0.251 (0.167 to 0.335)†† | 0.148 (0.059 to 0.237)†† | 0.106 (0.009 to 0.202)†† | 0.251 (0.167 to 0.335)†† | 0.172 (0.081 to 0.263)†† | 0.125 (0.029 to 0.221)†† |
Restricting drinks | 0.173 (0.110 to 0.236)†† | 0.138 (0.071 to 0.205)†† | 0.123 (0.021 to 0.226)†† | 0.173 (0.110 to 0.236)†† | 0.147 (0.083 to 0.211)†† | 0.112 (0.045 to 0.180)†† |
Daytime toilet trips | 0.198 (0.100 to 0.296)†† | 0.088 (−0.024 to 0.201) | 0.068 (−0.047 to 0.182) | 0.198 (0.100 to 0.296)†† | 0.087 (−0.024 to 0.198) | 0.057 (−0.056 to 0.171) |
Rewards | 0.281 (0.201 to 0.361)†† | 0.141 (0.049 to 0.233)†† | 0.088 (−0.014 to 0.191) | 0.281 (0.201 to 0.361)†† | 0.158 (0.062 to 0.254)†† | 0.143 (0.044 to 0.242)†† |
Showing displeasure | 0.054 (−0.091 to 0.199) | 0.009 (−0.146 to 0.164) | −0.052 (−0.214 to 0.110) | 0.054 (−0.091 to 0.199) | 0.009 (−0.134 to 0.151) | −0.047 (−0.187 to 0.093) |
Protection pants | 0.290 (0.182 to 0.398)†† | −0.010 (−0.161 to 0.141) | −0.010 (−0.161 to 0.141) | 0.290 (0.182 to 0.398)†† | 0.025 (−0.134 to 0.184) | 0.025 (−0.134 to 0.184) |
The estimates provided in this table are average treatment effects for each strategy. They are risk differences, that is, estimated differences between the risk of bedwetting at 9½ years after receiving given ‘treatment’ (ie, parental strategy) and the risk of bedwetting if they had remained ‘untreated’. We provide examples of how to interpret the risk differences below:
(1) The risk difference for ‘restricting drinks’ (0.123 (95% CI 0.021 to 0.226)) shows that there is a 12.3% (2.1% to 22.6%) increase in risk of bedwetting at 9½ years among children whose parents used restricting drinks compared with children whose parents did not use this strategy.
(2) The risk difference for ‘showing displeasure’ (−0.052 (−0.214 to 0.11)) shows that there is a 5.2% reduction in the risk of bedwetting at 9 years among children whose parents show displeasure, but the 95% CI indicates that this result could be in either direction (between a 21% reduced risk and 11% increased risk).
*Empty model column for propensity score-based methods analysis shows unadjusted differences in risk of bedwetting between ‘treated’ and ‘untreated’ children. These coefficients are equivalent to bivariate regressions including bedwetting at 9½ years as an outcome variable and each strategy as a single predictor.
†Model adjusted for child and family variables represents average treatment effects on treated children. These coefficients are estimated differences in risks of bedwetting in weighted samples. These are differences in risks adjusted for child and family variables that accounted for the differences between treated and untreated groups and between those with and without bedwetting at age 9½. The list of child and family variables was derived separately for every strategy on the basis of ORs from regression analyses providing evidence of an association (ORs <0.90 or >1.20) both with the given strategy and with bedwetting at age 9½. A detailed list of child and family variables included in the model for each parental strategy is provided in the online supplementary material (see online supplementary table S2–S7).
‡Model adjusted for child and family variables and other parental strategies. This is the column with coefficients adjusted for the child and family variables and for other parental strategies that were highly correlated (coefficient of tetrachoric correlation >0.45) with given strategy.
§Empty model: results of univariable logistic regression analyses including bedwetting at 9½ years as an outcome variable and the given strategy as a single predictor. To ensure direct comparisons with the output of the analysis using propensity score-based methods, the results of regression analyses are expressed here in terms of risk differences instead of ORs.
¶ Model adjusted for child and family variables. This includes estimated differences in risk of bedwetting for given strategy obtained from logistic regression, also including child and family variables.
**Model adjusted for child and family variables and other parental strategies. Risk differences for given strategy obtained from logistic regression including child and family variables and other strategies associated with the strategy under examination.
††Indicates risk differences that provide evidence for an increase in bedwetting associated with the parental strategy.