# Fixed effect omits non-fixed variables

I am working on a standard gravity model and am running a cross-section and panel regression which include fixed effects (country fixed effects for cross-section and country-time and country pair for panel). The results however omit some variables which are not "fixed" across time or country. There are 21 countries in the database. Any reason why that is so?

Edit: Update: I tried various combinations of exporter, importer fixed effects and sort of got through the problem of collinearity (I am using PPML command). However, some of my coefficients are as high as 43 now which is ridiculous.

• Without any of your code, data, results or anything it is impossible to tell. – Peter Flom Dec 7 '15 at 13:49
• I am not aware of how to include codes here so please pardon by poor formatting. However, my command was this: ppml trade GDPexp GDPimp PortDummy lnroad lndist diff lnsubs1345 lnsubsimp1345 ComBor ComLan exporter_* importer_* if year==2011, cluster(id_pair). As per regressions, the variable "diff" which is a log differential of prices between two regions, was found to be collinear with importer dummies (18, 19, 20 and 21). I would include the results here but I am not sure how to. Removing those dummies led to coefficient of "diff" as high as 43, compared to (-2) for other years. – Shaleen Khanal Dec 7 '15 at 14:51
• You can include code in your question by typing it in or cutting and pasting, then highlighting it and then clicking the "code" icon. (It looks like brackets) – Peter Flom Dec 7 '15 at 22:37
• ppml trade GDPexp GDPimp PortDummy lnroad lndist diff lnsubs1345 lnsubsimp1345 ComBor ComLan exporter_* importer_* if year==2011, cluster(id_pair) Ok. This is the code I am using. [drive.google.com/file/d/0BzsUh3A-jFZrbGlBamRZMnFTejA/…. The results look something like that. As you can see the "diff" variable is missing. After I remove importer_18-21 dummies, diff does appear but its coefficient reaches 43. – Shaleen Khanal Dec 8 '15 at 0:15