Housing, health, and the well-being of children

Methodology

We used the data collected within Growing Up in New Zealand, a child cohort study covering children antenatally until they were six years old. Pregnant women were eligible for the study if their estimated delivery date was between 25 April 2009 and 25 March 2010, and they resided within three contiguous District Health Board (DHB) regions in the northern part of the North Island (Auckland, Waikato, and Counties-Manukau DHBs). Morton et al. (2015) provide more detailed information about the study design and sample.

For this study, we used three waves of data to varying degrees. These included the following:

• Antenatal wave: survey of mother and mother’s partner primarily conducted in 2009/2010;
• 9-month wave: survey of mother and mother’s partner primarily conducted in 2010/2011, with the mother being the main proxy for questions about the child;
• 2-year wave: survey of mother, mother’s partner, and child’s proxy conducted primarily in 2011/2012. We primarily used information from the 9-month survey as that had the best measures of housing conditions for our analysis, allowed us to use the full sample, and was a time in the children’s lives when they tend to spend a substantial portion of their time in the home. We also used health outcome measures from the 2-year survey but were limited to only those children that did not move between the 9-month and 2-year surveys. This not only is a much smaller sample but is likely to be a selected sample of children. 

For this report, we focused on the unsafe and substandard housing conditions in the WHO Housing and health guidelines that are most relevant for New Zealand. For household crowding, our main measure of crowding is based on the number of household members per bedroom. To assess the prevalence of conditions pertaining to the interior environment of the home, we estimated the proportion of households that are cold, damp, or mouldy as well as the proportion of households where children are likely to be exposed to second-hand smoke either from the mother, the mother’s partner or others in the household. Using the indicators of houses being crowded, cold, damp, or mouldy, we
estimated the proportion of homes with these conditions. We also estimated the proportion of children whose bedrooms were reported as cold, damp, or mouldy. While we estimated these conditions separately, we know that they are often found together. 

After developing our indicators for these different categories of housing conditions, we assessed the frequency with which these different types of conditions occurred in the same homes. For example, we expect that more crowded homes are more likely to have problems with damp and mould, but we have not seen any research that looks at this in detail.

Given the number of questions related to some of these conditions and the relationship between the factors themselves, we used a principal components analysis to determine which variables are in fact measuring the same underlying factors and if indices can be developed (which are by design orthogonal to each other) as stronger measures for these conditions.

For the housing-related health problems, we looked at those health problems that have been previously associated with our poor housing conditions in the literature, and we have conducted an extensive literature review to determine
which health problems have been associated with these poor conditions. The health outcomes to be examined include the following:

• at 9 months: ear infection; chest infection, wheezing, bronchiolitis, bronchitis, asthma, pneumonia, croup; cough lasting for a week or more; gastroenteritis; skin infection;
• at 24 months: ear infections; chest infections, bronchiolitis, bronchitis, pneumonia, croup; cough in the last 4 weeks; wheezing in chest; and gastroenteritis. 

Using these health outcomes, we estimated the proportion of children with these conditions.

For binary dependent variables (e.g., whether the house is damp), we used logistic regressions with robust standard errors and coefficients reported as odds ratios. For categorical dependent variables counting number of events (e.g.,
number of times child had an illness, number of times child went to see the doctor), we used an ordered logistic regression with robust standard errors and coefficients reported as odds ratios. For counts of variables that are top-coded (e.g., number of respiratory hospitalisations, number of days in hospital for respiratory condition), we used a censored Poisson regression. For the continuous variables (e.g., housing condition index), we used Ordinary Least Squares with robust standard errors.

An odds ratio is a measure of association between an exposure (e.g., damp house) and an outcome (e.g., chest infection). The odds ratio is the chance that the outcome will occur given the exposure relative to the chance the outcome will occur in the absence of the exposure. The odds ratio is often written in terms of the number of cases (e.g., number of people with a chest infection) and non-cases (e.g., number of people without a chest infection) and their exposure (e.g., whether people live in damp houses). Hence, if the odds ratio is equal to one, the proportion of cases to non-cases is equal regardless of exposure. Similarly, if the odds ratio is greater than one, then the exposure increases the likelihood or risk of the outcome. Conversely, if the odds ratio is less than one, then the exposure reduces the likelihood or risk of the outcome. Hence, in the regression analyses, the null hypothesis is that the odds ratio is equal to one.