Yelp II: Same model, three cities, three different answers

The regression in Yelp I told us what predicts check-ins. The next question was where. Do restaurants in different cities behave differently?

We took 1,000 restaurants each from Las Vegas, Montreal, and Charlotte and ran an ANOVA.

One wrinkle first. The test assumes the three groups have similar variance, and ours didn't. Levene's test failed. Rather than ignore that, we switched to Welch's test, which corrects for it. Assumption checking isn't paperwork. It decides which results you can trust.

The results were clear. All three cities differed significantly from each other, not just Vegas from the rest. On the log scale, Las Vegas restaurants averaged the most check-ins, Charlotte sat in the middle, and Montreal came in lowest.

Why? At this stage, we couldn't say. Maybe Vegas restaurants are bigger and built for tourists. Maybe tourists check in more than locals do. ANOVA can confirm the cities differ. It cannot explain the difference.

But we already knew from Yelp I that review count was a powerful predictor. So the obvious question: are these city differences real, or are they just review-count differences wearing a disguise?

Answering that took one more model. That's Yelp III.