Missing pvalue, confidence interval and t test in regression results

[Giancarlo Carta]

Hi,

does anyone know why stata doesn't shoe confidence interval, t test and p-value in the results of the following regression?

xi reg dep var indevariables, robust cluster(pairid) where pairid indicates every possible combination of three string variables

Stata shows:

[Carlo Lazzaro]

Giancarlo:
are you sure that predictors with missing confidence interval, t test and p-value ar not constant indeed?

[Giancarlo Carta]

They are dummy variables which are constant over time, like Lang or BORDER

[Stephen Jenkins]

Have a look at the correlation matrix for your predictors. I suspect near-collinearity in some cases. The missing entries in your estimates table points to the inability to reliably invert the variance-covariance matrix associated with your OLS estimation procedure. As OLS is typically very robust, the problems you are having are likely to originate with your data or, related, the particular model specification. Once you've looked again at the relationships between your predictors, it'd probably be advisable to start with a relatively simple model (with few predictors), and successively add them -- this will also help isolate the problems you're having.

[Giancarlo Carta]

Well, I know where the problem is cause I have changed a variable and then this has occured.

I am estimating a model to predict trade.Originally for variable landlock I had two different variables:

-landlo_orig which has 0 value if the origin country has access to sea, 1 otherwise
-land_dest which works in the same way for destination country.

Following literature I have inserted a new variable landlock which has 0 value if in a pair of countries both have access to sea, 1 if one of them have, 2 if none of them have.

I have checked the correlation matrix and the data for this variable and everything seems ok.

So i really don't know what happens

[Stephen Jenkins]

(1) Crosstabulate landlock with Autonomy (a dummy variable?) and check whether there are obs in every cell. Also there is presumably no inter-temporal variation in landlock, and is this why the _cons estimate also doesn't have an SE reported? (I'm guessing now.) Whatever, a coefficient of -55 on Autonomy looks rather "suspicious", given its magnitude relative to the coefficients for other predictors and use of logged depvar.
(2) To me, it's rather weird to enter landlock as a single predictor (which can take on only values 0/1/2). Shouldn't you be entering landlock as a factor variable?