Question

One of my recent papers examine important and timely research questions using a field experiment approach in eBay auctions: (i) Can merchandise return policy (MRP; liberalness in the MRP) increase consumers’ willingness to pay? and (ii) is the marginal impact of MRP diminishing? In this study we created three brand new eBay seller profiles, all with zip-codes located within five miles of each other in a college town in the U.S. The eBay stores received exactly the same product description, pictures, outbound shipping policies, etc. The only difference among the three sellers was the extent of liberalness in the MRP and we chose to operationalize MRP liberalness in terms of the time window during which the customer is allowed to return the purchased product. The most conservative MRP (Storefront 1 and 1a) involved a 15-day return window. According to trade publications, this return condition is more conservative than retail-industry averages. Storefront2 and 2a received a 30-day return window, which corresponds closely with retail-industry averages. Finally, Storefront3 and 3a received a 60-day return window, which is more liberal than many retailers offer at this point. The other elements of the return remained constant across the three storefronts. Therefore, in terms of overall return-policy liberalness, it could be argued that Storefront1/1a < Storefront 2/2a < Storefront 3/3a . It is important to note that it is very common in my data that we observe a customer’s bidding behavior in several auctions.

[Question] During the revision stage of the journal publication process, one of reviewer’s comment was that the I may use a fixed effects model to control for unobserved individual fixed effects. Do you agree or disagree with the above statement? Please explain with details.

Answer 1

Previously, the researcher used the regression model to predict coupon redemption, based on the consumer's income level, while controlling for the variable size of the shopping basket and demographics regarding having children in the family. The regression model can be represented as follows :

y=β_{​​​​​​0} + β ₁ +X1+ β ₂ + X2+ β_{​​​​​​3} +X3+E.

Where:

x1 = consumer income
x2 = size of the shopping basket (control variable).
x3 = demographics regarding having children (control variable).
However, entering the fixed effect factor would provide a better estimate of the model, as it would improve the prediction based on the fixed effect suggested by the reviewers. for individual-level effects by using dummy variables, and therefore a more accurate prediction is possible.
The model would be better represented as follows:

y=β​​​​​​0 + β 1 +X1+ β 2 + X2+ β3x3+E+vi

Where Vi is the fixed effect of the whole factor.
Therefore, the best regression mode should take the combined effect of fixed and random effect.
In summary, I can say that I agree with the above statement..