Skip to main content

Table 3 Comparisons in the odds of shorter neighborhood tenure between models on data before and after the propensity-score-match

From: Residential mobility among adult cancer survivors in the United States

Before matching Crude odds ratio (95% CI) 0.40 (0.39–0.42)
Neighborhood Tenure Unweighted frequency Total (Cancer / non-Cancer) *Weighted % for Cancer / non-Cancer
Less than 1 year 25,568 (1279/ 24,289) 6.4 / 13.7
1–3 years 39,048 (2405/ 36,643) 12.5 / 21.7
4–10 years 46,057 (4028/ 42,029) 20.7 / 25.8
11–20 years 32,886 (3875/ 29,011) 21.2 / 19.2
More than 20 years 42,078 (7518/ 34,560) 39.2 / 19.6
After Matching Crude odds ratio (95% CI) 1.05 (1.047–1.06)
Neighborhood Tenure Unweighted frequency Total (Cancer / non-Cancer) *Weighted % for Cancer / non-Cancer
Less than 1 year 2130 (1091/ 1039) 5.9 / 5.6
1–3 years 4323 (2201/ 2122) 12.9 / 12.2
4–10 years 7350 (3692/ 3658) 21.0 / 20.9
11–20 years 7098 (3557/ 3541) 21.6 / 21.6
More than 20 years 13,617 (6718/ 6899) 38.6 / 39.8
  1. Notes: The crude odds of shorter neighborhood tenure was based on a cumulative logit model using the surveylogistic procedure, where the response variable was neighborhood tenure, which was a 5-level ordinal variable (Reference = More than 20 years), and the explanatory variable was cancer history status (Yes vs. No). We conducted propensity-score-match using a one-to-one nearest-neighbor matching algorithm that pairs participants with closest probability (caliper =0.25) of having a history of cancer, which were conditioned on the following 15 covariates: age, sex, race, Hispanic ethnicity, education, marital status, employment status, income, health insurance coverage status, residence region, birth place, family size, having family member aged 65 and older, having family member aged 18 and younger, and perceived neighborhood social cohesion, from the survey logistic model. We matched 17,259 participants, which was 90.3% of all 19,105 available cancer samples. We then applied the surveylogistic procedure to the matched samples. The standard differences of the 15 variables prior to and post- matching ranged from −0.5 to 1.1, and from − 0.04 to − 0.08, respectively (details in Table S2). *, row percent. The proportional odds assumption was met