0
$\begingroup$

I have a large data set of 700 ebay auctions and want to examine seller reputation effects on auction revenue. some sellers have "detailed ratings" (about 45%), these ratings are out of 5 stars across four categories. i have all the data on detailed ratings for those that have them but of course i do not (because as it doesn't exist) have data for about 55% of the observations. I was looking at various ways to deal with the missing data as i really dont want to loose all of those observations. I was thinking to generate a dummy "HASDETAILEDRATINGS" coded to 1 if a seller has detailed ratings and then include the results of the four detailed categories for those observations that have them. would this be valid and if so how would i interpret coefficient results?

PS. if anyone can suggest a better way to deal with this problem i would be most grateful. thank you!

$\endgroup$
1
$\begingroup$

For a few good ways to deal with missing data, this makes a great read:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3701793/#!po=4.60993

You do have a large chunk of data missing, so you may want to consider whether this data can be considered “missing at random” or if there is some underlying reason that this data is missing for some records. If you’ve only got the detailed ratings for those auction participants that have a certain level of experience, for example, it may not be reasonable to believe that missing values of the less experienced auction participants can be estimated or discussed with the data you have.

If you are working with data that has values missing at random, the multiple imputation, full information maximum likelihood, and expectation-maximization. Including any auxiliary variables beyond the single star rating will likely help with performance.

$\endgroup$
  • $\begingroup$ Yes, they are not just randomly missing. sellers need to have 10 detailed reviews before their detailed ratings are displayed. this is why i believe it would be incorrect to input a number (either zero or the variable mean) into all the blanks of these columns. With my current method (a seperate dummy column indicating whether a observation even has detailed reviews first, then missing data cells left blank) would i get biased results? also how would i interpret the coefficients? $\endgroup$ – Tom Witten Jul 24 '18 at 10:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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