Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

Suppose I am using a restaurant rating as an explanatory variable in a regression. The rating is defined as $R=\frac{G}{G+B+N+S}$, where $G$ is good, $B$ is bad, $N$ is neutral and $S$ is silent. I have two conceptual issues. First, I want to adjust the rating when it's based on a small number of experiences, perhaps by shrinking it towards the overall mean. Second, people seem overly reluctant to rate highly-attended restaurants ($S$ is very high for places like the local joint that's been around for 30 years). If I plot ratings against $\ln(\text{experiences})$, I get an inverted U shape.

Are there any transformations that I can use to remedy these two issue?

share|improve this question
2  
Have you explored directly the relationships between $S,B,N,M$ and the dependent variable? –  whuber Nov 27 '12 at 19:10
    
I fixed some notation above. Unfortunately I don't have all the data to do this, though I might be able to get it eventually. I have tried adding the full interaction between $R$ and the denominator of $R$, which is equivalent to adding $G$ and the denominator in the model. $G$ has a small positive effect and the denominator has a small negative effect. I also tried using the natural log of the denominator, with similar results. –  Dimitriy V. Masterov Nov 27 '12 at 19:37
    
What is your dependent variable? –  Peter Flom Nov 27 '12 at 20:16
    
The outcome is various measures of individual-level patronage (like check size, and number of meals). –  Dimitriy V. Masterov Nov 27 '12 at 21:39
    
@PeterFlom, I made transfromation a synonym for data-transformation, you can vote to approve the synonym here. –  gung Nov 29 '12 at 13:55

1 Answer 1

If the rating is fairly predictable based on the number of people who've been to the restaurant, it occurs to me that one possibility is to build a model of the ratings given experiences. Then you could use the residuals (obviously, constant variance is fairly crucial here) instead of the raw data.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.