Limitations

View the most recent version.

The potential for bias due to respondent attrition is problematic in longitudinal research.  From one National Population Health Survey (NPHS) cycle to the next, respondents were lost for reasons such as refusal to participate, item non-response, death, institutionalization, or relocation out of country.  Selective loss to follow-up, notably failure to collect information from respondents who had poorer mental health, may have weakened the observed association between marital dissolution and depression.  Out of the pooled total of 27,662 respondents assessed in the baseline cycles, 2,333 (8.4%) did not respond in the follow-up cycle.

The survey weights were those applied to the cycle 1 (1994/1995) data.  The weights were not adjusted to account for subsequent non-response, which could have biased the estimates if continuers in the longitudinal panel differed from non-respondents according to characteristics considered in this analysis.  To assess the potential effects of non-response on the results, the weighted proportions or averages of selected variables between non-respondents and respondents were compared.  Statistically significant differences by respondent status emerged:  non-respondents were more likely than respondents to be male and to have been in a common-law relationship (rather than legally married) at baseline (data not shown).  Because of these differences, it is possible that the strength and/or magnitude of the observed associations between marital dissolution and depression in the study population were different from what they might have been if non-respondents had participated.

Information about whether respondents had ever been professionally diagnosed with depression was collected at cycle 6 only, and appended to the earlier records for the same respondent.  Owing to loss to follow-up, the likelihood that this information could be appended to respondents varied by cycle:  it was missing from less than 1% of records for cycle 6 (2004/2005), but 23% of records for cycle 1 (1994/1995) (data not shown). 
 
Because NPHS interviews are conducted every two years and respondents are asked about depressive symptoms that occurred during the 12 months before the interview, those who experienced depression only in the year following their baseline interview are categorized as not having suffered from depression.  Misclassifying respondents because of the one-year period would dilute the strength of the true association between marital dissolution and depression.

The NPHS does not ask respondents when marital dissolution occurred.  It could be argued that knowing the precise date would not contribute substantively to the analysis, since marital breakdown is an extended process that does not occur at a single point in time.  On the other hand, the point in the process at which the interview took place could influence the findings.  It is also unclear if depression preceded the breakdown of the relationship or if depression followed the marital dissolution. 

The definition of “remained married” required only that respondents report that they were married both at baseline and follow-up two years later, even though they might have divorced and married someone else.  Such a scenario was probably rare, given the frequency of data collection (every two years).

Sample size restrictions prevented analysis of associations between the dissolution of specific types of marriages (legal versus common-law) and subsequent depression.

Household income is based on total household income, household and community size and the associated low-income cutoff  (LICO).  While child support and alimony are included in the calculation of total household income, such payments are not deducted from the household incomes of those who pay them.  As a result, the adjusted household income of those paying support might be overestimated.

The NPHS data show an association between marital dissolution and depression; causality cannot be inferred.  As well, the associations observed may result from factors not considered in this analysis such as beliefs about marriage, spousal infidelity, re-marriage, age at marriage, marriage duration, or which partner initiated the break-up. 

The analyses are based on self-reported data, and the extent to which the data are biased because of reporting error is unknown.