Wealth does indeed predict undeliverable mail ballots, but received ballots problems are in fact negatively correlated with wealth

In my post on the 1,248 Philadelphia mail in ballots with problems, I ran two analyses of how ballot deficiencies track with income, and both came out surprising. The paperwork failures (e.g., bad signatures, and unverified IDs, and missing envelopes) had no income relationship at all, which runs counter to intuition that poorer people fare worse in complying with bureaucratic requirements. The second finding was also counterintuitive, namely that undeliverable ballots were more prevalant amongst richer ZIPs - an unexpected result, since housing instability is higher amongst poorer zips, suggesting USPS should hav emore challenges to deliver ballots in poorer areas. .

Both my analyses were by ZIP, on per capita rates. I knew per capita was an imperfect denominator, because there are markedly different voter turnout rates between richer and poorer areas (about a 2x differential in Philadelphia politics). However, the perfect data (by zip, real-time, for the Nov. 2026 general election) did not exist so ran some light proxy adjustments offline, demonstrating that, at least my higher income → more undeliverability was not explained away.

A friend pointed out that was hand-wavy (a fair point), so I decided to do a more formal version of the analysis with the data that is available, namely May 2026 primary election voter stats by ward. Note that this is from a different election 6 months ago, and is by ward instead of by zip. That's why its imperfect, but given voting patterns and income demographics dont change markedly over 6 months, this is a pretty good instrument variable and, frankly, the best that exists to normalize the current general election data.

For the May 2026 primrary in Philadelphia, there were 241, 126 ballots cast, of which 67,289 were mail-in ballots. The aggregate turn out was 22.3% of registered voters, and about 27.9% of all cast ballots were mail-in ballots. This data exists by zip, not ward, so I recast all of the below analysis by wards, which the current City Commissioner problematic mail-in ballot data does allow.

Of the 279 received mail-in ballots with problems, adjusting for voter rates does reveal a negative correlation

So this analysis changes my original result significantly. Now there is a clear negative correlation between higher income and lower ballot deficiencies, consistent with expectations. Said another way, poorer people have more trouble filling out or complying with government ballot form-filling policies. More specifically, a family earning $30k/year has roughly 3x a chance of their ballot being rejected due to bad signatures or outdated ID than a family making $90k/year. Although in all fairness, the absolute chance of that remains low in absolute terms, about 0.7% or 1 per 135 mail in ballots received overall or about 2% for poorer voters and 0.6% for richer voters.

However, the relationship between higher income and higher undeliverability remains, even after adjusting for voting rates

The ballots that never arrived behave better. Raw count r = 0.61, per capita r = 0.67, per mail ballot received r = 0.31 and still significant at p = 0.011. Weaker than the first post said, but the same direction.

The eight richest wards average 6.14 undeliverable ballots per 1,000 ballots cast and the eight poorest average 2.55. My original post calculated a 3.6x worse raw undeliverability rate, before adjusting for the 2x higher voter turnout rates in richer areas, meaning about a 1.8x effective rate. Using this smarter analysis by ward, I was able to calculate a more precise value of 2.4x. In other words, richer areas, controlling for higher mail in ballot voting rates, have about 2.4x higher chance of having undeliverability issues. What's fascinating here is that the top 6 richest wards are pretty diverse in terms of age, which is the primary explanatory variable, since young people change addresses more often. All that being said, the core novel result from my original work holds true: for the current Philadelphia general election, there is a significantly higher undeliverability rate for richer areas than lower areas, irrespective of whether you measure per capita or by mail-in ballot rates.

Notes, Sources, and Methodology

Deficiency data is the City Commissioner’s 2026 deficiency list, last updated Sept. 30, 2026, the same file as the first post. The 969 undeliverable and 279 paperwork split follows the Pennsylvania Department of State’s mail-in ballot guidance v3.4.1, page 7.

Ward turnout is the City Commissioners’ voter turnout details for the May 19, 2026 primary, all 66 wards reporting, pulled Oct. 5, 2026. The Pennsylvania Department of State’s own primary mail ballot request dataset agrees closely (68,474 ballots returned against the turnout page’s 67,289 mail votes), but it puts every Philadelphia row at a single county centroid, so ballots sent and returned cannot be resolved by ward from it either. That is the whole reason the sent-out panel is empty.

Ward income is ACS 2023 5-year block groups area-aggregated to ward boundaries, and is a household-weighted mean of block group medians, an index for ranking wards rather than a quoteable ward median.

Correlations are Pearson r with two-tailed p-values, and Spearman agrees on direction everywhere quoted here. Counts are small, a median ward has 13 deficiencies, so the count models use a Poisson GLM with each denominator as an offset.

Ward income predicts mail volume (r = +0.65), so dividing by mail ballots can manufacture a correlation out of an unrelated numerator. Simulating 400 draws of paperwork counts drawn independently of income, with the mean proportional to May mail volume, gives r = +0.01, 95% band -0.18 to +0.22, against an observed -0.51. The relationship is in the counts. The undeliverable rate, by contrast, correlates with its own denominator at r = -0.05.