[Joao Santos Silva]

Dear Guest,

The interpretation is as in a linear model with a dependent variable in logs: for regressors in logs the coefficient is an elasticity (measured in %) for regressors not logged, the coefficient is approximately a semi-elasticity (in p.p.).

Best wishes,

Joao

[Dr. Iqra Yaseen]

following

[Dr. Iqra Yaseen]

Guest, is it advisable to use log variable of GDP and DIST when using ppmlhdfe command. Joao Santos Silva

[Guest]

Dear Guest,

The interpretation is as in a linear model with a dependent variable in logs: for regressors in logs the coefficient is an elasticity (measured in %) for regressors not logged, the coefficient is approximately a semi-elasticity (in p.p.).

Best wishes,

Joao

Professor, regarding not logged coefficient for NTM (non-tariff measures), NTM_t = NTM_t + NTM _t-1 (applied in cummulative) the coefficient is - 0.0128. In this regard, since it is the semi-elasticity can I applied (100*(exp(beta)-1)) formula for interpretation. Second NTM as in restrictiveness (frequency index (%) ) which is not logged as well. The coefficient in this analysis is - 0.0063. Can I apply the previous formula to interpret this outcome too?
Thanks again for your time and consideration.

[Guest]

sybil arqi
Please visit this forum, "Problem with generating country pair fixed effects in Gravity estimation". (To answer your question, "Yes") Professor Santos Silva please correct me if I am mistaken it.

[Dr. Iqra Yaseen]

Joao Santos Silva kindly clarify whether to generate log DIST and log GDP while using ppmlhdfe.....or to use them in their original form?

[Joao Santos Silva]

Dear sybil arqi,

GDP and distance should indeed be logged.

Best wishes,

Joao

[Joao Santos Silva]

Dear Guest,

That is correct, but with estimates so close to zero the transformation will have little effect and the result will be almost identical to the estimate itself.

Best wishes,

Joao

[Guest]

Professor, thank you very much.

[Dr. Iqra Yaseen]

Joao Santos Silva fine. i have imports and exports as dependent variable and rta, common language, contiguity, common colony , population, GDP as independent variables. may you please tell me which one log and which ones to not log as i am using ppmlhdfe command. thanks

[Joao Santos Silva]

Dear sybil arqi

Log all the explanatory variables that are not dummies. Do not log the dependent variable.

Best wishes,

Joao

[Ridwan Sheikh]

Dear Joao Santos Silva .
My regressor is an index (contineous variable) that lies between 0 to 5 and i am taking its logarithm. Suppose, I am estimating an exponential mean model y=exp(X)n_ij,t using PPML, where y is trade flows (levels) and X is a vector of covariates. My coefficient on logarithm of index is 1.25. Can i interpret it as, 1% increase in index leads to almost 1.25 % increase in trade flows.

Thanks,
(Ridwan)

[Joao Santos Silva]

Dear Ridwan Sheikh

Assuming the index has no zeros and has cardinal meaning (i.e., 4 is 2 times 2, not just larger than 2), then that is correct.

Best wishes,

Joao

[Ridwan Sheikh]

Thanks Joao Santos Silva , this was really helpful.
It is possible to have an index value of Zero, If that be a case, then we should avoid log-transformation and use the regressor (said index) in levels and interpret the coefficient as semi-elasticity. Is that what you mean?
Thanks,
(Ridwan)

[Joao Santos Silva]

Dear Ridwan Sheikh,

Yes, if you have zeros you cannot take logs.

Best wishes,

Joao