Yelp III: Are cities really different, or do they just have more reviews?

Yelp I found that reviews drive check-ins. Yelp II found that cities differ. Yelp III asked whether those two findings are secretly the same finding.

We ran a General Linear Model with city as the independent variable, review count as a covariate, and the interaction between them. In plain terms: we asked whether the city gaps survive once you account for how many reviews restaurants in each city have.

The honest answer required some statistical care. Our residuals weren't normally distributed and weren't homoscedastic, so we used bootstrapped confidence intervals instead of relying on the standard ones. And because the effect of reviews differed by city (the interaction was significant), a plain ANCOVA was off the table. The GLM handled it.

Here's the result that makes the whole three-part series worth it. Before controlling for reviews, Las Vegas beat Montreal by 1.88 on the log check-in scale. After controlling, the gap shrank to 0.54. Vegas over Charlotte went from 0.66 to 0.15. The differences stayed statistically significant, but they shrank dramatically.

Translation: much of what looked like "city culture" was really review volume. But not all of it. Something about the cities themselves still differed, and with this dataset, we couldn't say what. Knowing where your data stops being able to answer questions is part of the analysis too.

The full arc across three models: reviews, open status, price, and delivery best explain check-ins. Cities genuinely differ. And a difference that looks huge in one model can shrink to modest once you control for the right variable. That last lesson applies to nearly every dissertation dataset I've seen