May I be permitted to ask this again in the hope of obtaining an answer? Clearly I'm missing something obvious, but how does one do this? Correlating the variable with the model residuals to assess this is, I believe, incorrect. I consulted the guide and turned up dry, ditto the command reference. Running the coefficient covariance matrix doesn't include the residuals. I've also run a Google search and could only find this: http://lists.wfu.edu/pipermail/gretl-users/2011-July/006470.html which isn't quite what I'm looking for. This should be straightforward, so your assistance would be very much appreciated. -- Clive Nicholas "My colleagues in the social sciences talk a great deal about methodology. I prefer to call it style." -- Freeman J. Dyson

Sven, Check out IV estimation with the 'tsls' command. May I make the Running the panel diagnostics informed me that Residual variance: 1.35107e+06/(822 - 60) = 1773.06 Joint significance of differing group means: F(46, 762) = -1.99828e-14 with p-value 1.79769e+308 (A low p-value counts against the null hypothesis that the pooled OLS model is adequate, in favor of the fixed effects alternative.) but neither is a fixed effects model, because I'm actually fitting this as a multilevel model, which -gretl-, of course, cannot yet do (no criticism intended; I can do all that in R). I can't properly run IV2SLS models in -gretl- because I only have one IV and the main model has 96 predictors, so it responds that I don't have enough instruments. But I only have _one_ variable in that 96 which I suspect of being endogenous. -gretl-, however, does not allow me to identify that variable as (potentially) endogenous. Taken on its own, that's not an issue: again, I can run an IV model to my satisfaction in R, but the point is _I don't know for sure that this variable *really is endogenous* unless I have a suitable variable or variables as candidates._ If it really is endogenous, then I have no option to abandon the multilevel approach and switch to IV in the hope I can get enough suitable instruments. If not, I can proceed with the multilevel approach. So, what to do in order to test the endogeneity of my regressor in a 'standard' model? -- Clive Nicholas "My colleagues in the social sciences talk a great deal about methodology. I prefer to call it style." -- Freeman J. Dyson

Allin, v1.9.12. I'm not able to build the latest version of -gretl- on my Linux machine (you may remember the fun and games that ensued last time I tried this earlier this year), so I use the latest version that's made available from the repository. I consulted the command reference for -tsls- and it displayed a rather different example (one where the endogs were completely different to the exogs), but no matter: let me try that when I'm on my machine and I'll report back to you. I've been using point-and-click to run IV2SLS on -gretl-, so it would do me good to run it as syntax. Okay, thanks Allin. Over and out. :)

Okay, running -tsls- 'works' on my data and produces the post-estimation tests which allows me . Thanks for your assistance on this. I do have a conceptual question about assessing and selecting suitable instruments, however. When assessing them in running OLS models, I'm often finding that at least one of them shows up as significant, but if my reading of IV is right, a 'suitable' instrument cannot have its own independent effect on the response, only via its correlation with the endogenous variable. Correct? Otherwise, I'm left with just one variable which passes, but which doesn't quite make F > 10 on the weak IV test and no Sargan calculation (tough, you may say). On 13 November 2013 08:01, Riccardo (Jack) Lucchetti <r.lucchetti(a)univpm.it&gt;wrote: What details do you need? I'm rather scratchy on the technical details now. -- Clive Nicholas "My colleagues in the social sciences talk a great deal about methodology. I prefer to call it style." -- Freeman J. Dyson