Many studies have found that patients with lower socioeconomic status (SES) have worse access to care. For instance, patients from lower SES have longer wait times (see review papers Landi et al. 2018 and Siciliani 2016). One key question is whether these disparities depend on how socioeconomic status is measured: at the individual level or aggregated into geographic level?

A paper by Carlsen et al. (2024) aims to answer this question. The authors used data from two Norwegian data sets: (i) the National Patient Registry (NPR), which captures nonemergency hospital treatment and (ii) Statistics Norway, which captures SES. The authors outcome of interest is wait times (measured as the “number of days between the hospital receives the letter of referral and the patient is admitted to treatment”) with the key dependent variable being either SES defined at the (i)
individual, (ii) population cells (i.e., combination of gender, age, year and the municipality of residence) or (iii) municipality. This approach aims to test if there is aggregation bias when using geographic measures of SES compared to individual ones. Using this approach, the authors find that:

…the [waiting time] coefficients [for SES] are higher when socioeconomic status is measured at an aggregated level. For instance, the associations of tertiary education at the municipal level for females (males) are approximately 3.7 (3.3) times higher compared to the individual level. Regarding waiting time, these differences translate to an estimated disparity of 28 (21) days…

When socioeconomic status is measured at the individual level, we find a 7%–8% reduction in waiting time for those with highest educational status relative to those with lowest education.

One might think that these results are driven by the fact that hospitals in lower SES areas may be lower quality/have longer wait times, but the authors include hospital fixed effects. Another reasons could be that social networks could be important. A ‘middle class’ individual living in a richer area may have a social network with more privileged peers who can ‘work’ they system; putting that same person in a low SES region would make them less likely to have such an advantaged network. Conversely, individual level factors (e.g., education, resources, ability to speak the language) may be important as well, but could be exacerbated when low SES individuals are surrounded by similarly low SES individuals. This is just speculative, but the finding that the SES access gradient varies depending how SES is measured (i.e., individual level or geographic) is an important finding.