this research is not my field of science. so I can't answer the question

It is expected that with the standard error of 3.33E-08, a value of TRADE_BALANCE of -7.46E-08 with the t statistic of -2.242 is statistically significant. The result is OK.

Dear Lul Dol

Well, you should first of all look at the units in which you have expressed trade balance. I its quite possible that you have expressed it in very small units (e.g., US$) and it could be more appropiate to express such type of number in bigger ones (e.g., thousand US$). Therefore, first, put the numbers in a unit that can make more sense. In order to guide your decision, look firstly at the descriptive statistics to see which the mean and sd are.

In the second place, you look at the statistical significance and you can see and comment on the fact that you can reject that the estimated coefficiente is equal to 0.

Finally, assumming that you have the "right" units to interpret your result, you should comment on the social or economic relevance of your estimate. Is an effect of 0.001 due to a 1-unit increase of the variable of interest a relevant effect from an economic point of view? It is possible that is not and you should comment on it. Statistical significance and economic significance are different issues. Sometimes, in order to assess the latter, people put the estimates as a multiple of the standard deviation of the variable (standardized coefficientes); you could try it as well.

I hope it helps.

José-Ignacio

Thanks for all questions and answers above.

The value of my coefficient has the same issues/questions.

I tired the value of for the cost like like very small units or a bigger one. But, the results changed nothing.

I am also considering how to explain for this issue.

You are welcome. If solved, it is good you mark my answer as solved!

Dear Lul Dol

I am using the "Conditional Logit Model" in R-studio.

The value of my coefficients and p-values for some variables makes me a little bit confused about my result to explain it.

Anyhow, thank you so much for your kind suggestions.

20 observations are not enough. Otherwise you are a victim of micronumerosity, that your data set is inadequate hence the outcome shown through the screen shot is spurious

in this table results are ok and i think these results will reflect the right picture of your designed hypotheses. no need you confuse for interpretation bcz the trend of results is ok.

It is a unit of measurement issue, and it really doesn't matter in terms of interpretation.

s.s Somra Thank you very much, Sir. Yes, the result reflects my designed hypothesis. However, the value of the coefficient didn't come out as the only number. It includes e (10 to the power). I am wondering why the program (R-studio) produces my results like that. Still trying to understand it.

Lul Dol -

I suggest you use a "graphical residual analysis" when selecting your model, and use log transformations sparingly. You might want a quadratic linear regression. A graphical residual analysis will show your model fit to your data (including heteroscedasticity), but to avoid fitting too closely to your sample, and not so closely to the rest of the population, saving some data for "cross-validation," when you can, is advisable.

A different set of predictors might be advisable, with or without an intercept term, depending upon whether there should be a zero or nonzero y-value when all predictors are zero.

Comparing different models, using graphical residual analyses and cross-validations, could be helpful. You could compare everything from completely different models, to ones differing only by the inclusion or exclusion of one particular predictor..

Cheers - Jim

PS - Regarding logs, they sometimes have a special use in Econometrics. (I just noticed the word "elasticity" above.) However, I have noticed that many people just use them to reduce heteroscedasticity, a natural feature which I think should more often be considered in the error structure where it belongs. Overuse of logs can just make interpretation more nebulous, though I do not know that that is the case here, as I am not familiar with the subject matter.

You might want to look at the following: https://www.researchgate.net/project/OLS-Regression-Should-Not-Be-a-Default-for-WLS-Regression, and the various updates there, in reverse chronological order.