Continuing our theme of the week of data analysis related to performance, today we look at data from studies about restaurant reviews and their relationship to the weather. Data scientists at Georgia Tech, in conjunction with Yahoo Labs (@YahooLabs), studied 1.1 million reviews of 840,000 restaurants across the country. Similar to the Rate My Professor website I discussed earlier this week, this analysis looks for statistical correlations between positive and negative reviews of restaurants.
In this interesting paper, authors Saeideh Bakhshi, Partha Kanuparthy, and Eric Gilbert ask if weather and local demographics of restaurants drive how we rate them online? The authors find in their study that in addition to factors such as meal, price, and service--which logically one expects to influence restaurant reviews--other factors such as demographics of the local neighborhoods, snowfall, rain, and temperature--also seem to influence whether a review was postive or negative. In particular:
"Weather conditions are significantly associated with ratings. Reviews written on warm or cool days are more likely to be rated high than those written in cold or hot days. Reviews written on rainy or snowy days tend to have lower ratings than those written on days without rain or snow."
Similar to the Rate my Professor discussion, the real question still to be addressed is how do we interpret these types of statistical relationships we can find in large quantities of data, and what are the underlying statistical techniques one might use? Our next blog will address this issue.
The initial article brought this study to my attention can be found on the Weather Channel's Forecast Factor blog.