DRAFT
Given the large discrepancies in participation rates these should be treated with caution,
particularly in judging the performance of the BreastScreen program. Despite the
differences between self-reported and administrative data, survey data may still be valid
in analysing participation if no systematic variation in over (or under reporting) exists
amongst sub-populations. A study by Zapka et al (1996) which compared self-reported
mammography use with program data found no biases in self-reporting accuracy amongst
women of various ages, income or level of education. Whilst no such tests can be
undertaken here, the results from Zapka et al (1996) lends support to the use of survey
data for the purpose of analysing the distribution of participation.
Information on household income was missing for 22% of the sample, similar to the
proportion missing in the overall NSW Health Survey. Missing income data is not
unusual for household surveys that are not specifically designed to elicit information on
income or wealth (Doiron, Jones, & Savage, 2008). However, to avoid selection bias, we
included a dummy variable in the analysis for women who did not report their household
income. To check the potential influence of measurement error, we re-estimated the
model excluding observations with missing income. None of the income coefficients
changed significantly.
We found evidence that income was negatively related to the likelihood of a woman
screening regularly. In other words, women in higher income households were more
likely to screen regularly. A similar trend amongst the never screeners was found but did
not reach statistical significance except for the $20,000 to $40,000 income group, who
were less likely to have never screened compared to women in the lowest income group
(<$10,000). These results are consistent with Birch (2007) and the AIHW (2007)
although neither study report result which separate late screeners from never screeners in
their analysis. However, the results are at odds with Taylor et al (2003) who found that
women with household income greater than $40,000 were significantly more likely to
belong to either the never screened or_irregular screeners groups. The discrepancy may
be explained by the use of only two categorical income variables (less than or greater
10