[Clyde Schechter]

Scanning the output, it looks to me like you have a problem with iv12. That coefficient and standard error look unreasonable. It may be that iv12 has extremely low variation or something like that. Take a look at the distribution of iv12 itself and also a scatterplot of dv with iv12 to see if something is pathological there.

I suspect that things will not work until iv12 is either eliminated from the model or corrected.

[Nae Khar]

I ran it without iv12 and the F statistic showed up! Thank you!!! I have been struggling with this issue for weeks before finding this form. I also logging the variable and adding it again in the regression and it worked with the logged version of iv12.



iv12 was one of my controls. It represents GDP (PPP). Its a measure of GDP over a number of countries where an international dollar has the same purchasing power over GDP as a U.S. dollar has in the United States. I choose it rather than nominal GDP because it makes country comparisons more useful.


I attached a distribution of the variable GDP(PPP) as well as a scatterplot. The dv (startups) is the percent of business startups per year. This is what it looked like before logging.


How trustworthy are the confidence intervals in the fixed effects regression? If I don’t find any zeros in my confidence intervals (although some variables are showing to be significant), is that something I should be concerned about?

[Nae Khar]

This is my output after the fix:


xtreg dv iv1 iv2 iv3 iv4 iv5 iv6 iv7 iv8 iv9 iv10 iv11 iv12_ln iv13 iv14 iv15 iv16 iv17 i.year, cluster(code) fe





Fixed-effects (within) regression Number of obs = 372

Group variable: code Number of groups = 78




R-sq: within = 0.1919 Obs per group: min = 1

between = 0.1663 avg = 4.8

overall = 0.1681 max = 9




F(24,77) = 4.08

corr(u_i, Xb) = -0.1258 Prob > F = 0.0000




(Std. Err. adjusted for 78 clusters in code)

--------------------------------------------------------------------------------------------------

| Robust

startup | Coef. Std. Err. t P>|t| [95% Conf. Interval]

---------------------------------+----------------------------------------------------------------

iv1 | -.0491427 .0605414 -0.81 0.419 -.1696961 .0714107

iv2 | .1420266 .3000548 0.47 0.637 -.4554589 .7395121

iv3 | -2.141677 1.044211 -2.05 0.044 -4.220967 -.0623876

iv4 | .1106859 .0376451 2.94 0.004 .035725 .1856469

iv5 | -.0030259 .0180383 -0.17 0.867 -.0389447 .032893

iv6 | 1.065301 .3384834 3.15 0.002 .3912949 1.739308

iv7 | .1076039 .9619924 0.11 0.911 -1.807968 2.023175

iv8 | -.8422744 .4628608 -1.82 0.073 -1.763948 .0793991

iv9 | 1.775608 .7426309 2.39 0.019 .296841 3.254375

iv10 | .2998351 .4073326 0.74 0.464 -.5112676 1.110938

iv11 | -.3999043 .7755373 -0.52 0.608 -1.944196 1.144388

iv12_ln | .4162268 2.779559 0.15 0.881 -5.118582 5.951036

iv13 | -.1724294 .0787467 -2.19 0.032 -.3292341 -.0156248

iv14 | -.4848558 1.759241 -0.28 0.784 -3.987952 3.01824

iv15 | 1.457746 1.383883 1.05 0.295 -1.297917 4.21341

iv16 | -2.791635 2.626027 -1.06 0.291 -8.020723 2.437453

|

year |

2007 | -.1487876 .7654087 -0.19 0.846 -1.672911 1.375336

2008 | .6448665 .7886044 0.82 0.416 -.9254454 2.215179

2009 | .4651307 .8944082 0.52 0.605 -1.315864 2.246125

2010 | -.2608271 .9438673 -0.28 0.783 -2.140307 1.618653

2011 | 1.791091 .930258 1.93 0.058 -.0612892 3.643472

2012 | 2.156847 .9500815 2.27 0.026 .2649926 4.048701

2013 | 2.66559 .9715189 2.74 0.008 .731049 4.600131

2014 | 2.539329 .9316987 2.73 0.008 .6840795 4.394578

|

_cons | -2.092358 72.9084 -0.03 0.977 -147.2715 143.0868

---------------------------------+----------------------------------------------------------------

sigma_u | 6.9152562

sigma_e | 2.5613787

rho | .87935843 (fraction of variance due to u_i)

--------------------------------------------------------------------------------------------------

[Clyde Schechter]

How trustworthy are the confidence intervals in the fixed effects regression? If I don’t find any zeros in my confidence intervals (although some variables are showing to be significant), is that something I should be concerned about?

I don't understand this question. As I look at your output, the confidence intervals exclude 0 when and only when p < 0.05, which is exactly what one expects from test-based confidence intervals. What precisely are you concerned about here? What were you expecting to see that you did not. (Or what did you see that surprised you?)

[Nae Khar]

Sorry perhaps I was not clear on that.

I meant more specifically, if you look at iv8 there is a 0 in the confidence interval although it is significant (not at the 5% level, but at the 10% level). What should I make of that when I come to interpret that specific variable in the results of this analysis?

[Clyde Schechter]

Make nothing of it. As the output clearly tells you, the confidence intervals shown are 95% confidence intervals, so they will exclude 0 when and only when the corresponding p-value is less than 0.05. The intervals that would exclude 0 when and only when p < 0.10 would be 90% confidence intervals. If you want those, re-run the regression adding the -level(90)- option.

[Nae Khar]

Many thanks for all the clarifications. I highly appreciate it.

[Aye Aye Khaine]

Thank you all for the inquires and the expert responses!
I got what I was looking for!
Aye Aye Khaine

[sony vebiyanto]

Hi Carlo Lazzaro
I need some explanation from you about testparm command
I used the testparhm command to find the F-value after estimating PPML because the PPML estimation result did not display the F-value and it worked, is the interpretation the same as the F-value in OLS??

or is there a more appropriate way to find the F-value in the PPML estimator to determine the effect of all independent variables on the dependent variable simultaneously

[Carlo Lazzaro]

Sony:
via -parmtest- you test the joint significance of the levels a categorical variable is composed of.