To test the statistical significance of interactions between the environmental exposures and the XRCC1 genotype, we compared the models that included terms for all combinations of the XRCC1 genotype and levels of environmental exposure to the models with indicator variables for the main effects only (nominal LRT).
To test statistical interactions between the genotypes and cigarette smoking in unconditional models, we first used a likelihood-ratio test (LRT) to compare nested models that included terms for all combinations of the genotypes and cigarette smoking status to the models with indicator variables for the main effects only (nominal LRT).
The models with indicators for different cutoff points returned similar estimates for temperature, relative humidity and vapor pressure (Additional file 1: Figure S2).
Qualitative matching is incorporated into the model with indicator variables, i.e., some is an indicator variable for respondents in a "somewhat related" job, while close is an indicator variable for respondents in a "closely related" job.
If the continuous model was not significantly different from the model with indicator variables (p ≥0.05), the linearity hypothesis was accepted.
Owing to computational limitations, models of indicator variability (both fixed and random effects) failed to converge, hence we used a three level model with indicator observations nested within practices and practices nested within primary care trusts.
The number of correct turns, scored over five trials, was considered as count data for each time point during the experiment, with a random-intercept multilevel Poisson model with indicator variables fitted for each time point.
We also fitted a model with indicator variables for the deciles of O3[24hr] exposure averaged over the entire pregnancy and presented the difference in birth weight associated with each decile of O3[24hr] exposure relative to the lowest decile.
To obtain a summary stratified IRR, we used a fixed-effects M-H type procedure as implemented in the "ir" command in STATA, which should give results similar to fitting a Poisson regression model with indicator variables for "study".
We examined the dose-response association visually by fitting a regression model with indicator variables for categories of cumulative dose and plotting the resultant relative rate estimates against category specific, mean dose values.
To account for this nonlinear association and to evaluate the upper and lower extremes of Mn levels, we fit a multivariable regression model with indicator variables for the lowest and highest quartiles of Mn (hair and blood) with the reference group set to the middle two quartiles of Mn distribution (i.e., within the 25th to 75th percentiles) (Table 2).
