How to get T-ratio instead of Z's in PPML regression

[Cynthia Hourizene]

Dear all,

I am using Stata 13.0, I am estimating a gravity model using PPML, I get Z statistics on the output but I need to report T-stats. instead?

is there any formulas or option commands I can add to my PPML estimation or post estimation commands in order for me to get t-statistics instead?

I can post my code if needed

[Cynthia Hourizene]

I can post my code if needed

sorry I meant I can post my output if needed

, here is my code

ppml TradeValue LaggedTrade NAFTAmm EUmn ASEANm MERCOSURm lnExpGDP lnImpGDP lnExpPop lnImpPop lnExpProd lnImpProd lnFarmInc lnDistance BorderDummy qdatedum* Exporter_FE* Importer_FE*, cluster(lnDistance)

[Phil Bromiley]

You didn't get a quick answer. ppml is user written. As I understand it, the calculation of the score is the same - the difference is whether you treat it as z or t distribution. With N above 30 or so, t and z are pretty much the same.

If you really must have probably based on t I can immediately see two alternatives. First, you can open ppml.ado and look at what is calculating the z probability. You should be able to change that calculation line to a t. Alternatively, you can use the output of ppml to calculate your t's.

[Cynthia Hourizene]

Thank you Phil; When you say "open ppml.ado", does it mean I should type on Stata "ppml.ado" or there is something else I should do? I am sorry I am still learning to use Stata.

[Maarten Buis]

You cannot get t-statistics with ppml. The t-statitistics only occur in linear regression estimated with ordinary least squares. The Poisson model (estimated with ppml) is estimated with maximum likelihood, so the appropriate statistics are z-statistics. As Bill said, the computation is exactly the same. The only difference is the distribution with which these statistics are compared. Moreover, you typically don't do the comparing yourself, you let Stata do it by looking at the p-values that are returned on the same row in the output.

[Cynthia Hourizene]

Thank you very much Maarten