areg command not working

[Payal Seth]

Hello All,
I tried running a simple command:
reg haz06 odf i.village, and got the results (attached):


haz06 Coef. Std. Err. t P>t [95% Conf. Interval]

odf .6266667 .3014252 2.08 0.038 .0349016 1.218432

village
Pachrukhi .1234444 .2756082 0.45 0.654 -.417636 .6645249
Gerui -.1427083 .2714184 -0.53 0.599 -.6755634 .3901467
Nagwa .7018452 .2622151 2.68 0.008 .1870584 1.216632
Hathigarwa -.2654386 .287602 -0.92 0.356 -.8300656 .2991884
Itowa -.0629487 .3063975 -0.21 0.837 -.6644755 .5385781
Talbhirona .0541257 .2575683 0.21 0.834 -.4515384 .5597898
Puraina -.0127778 .2970486 -0.04 0.966 -.5959505 .570395
Marukharkala -.1656 .2908106 -0.57 0.569 -.7365263 .4053263
Harni Bujurg -.0864966 .2701229 -0.32 0.749 -.6168082 .443815
Sonversa -.2871088 .2701229 -1.06 0.288 -.8174204 .2432027
Rampurwa .0075806 .3237262 0.02 0.981 -.6279662 .6431274
Soherwalia -.1991026 .3205579 -0.62 0.535 -.8284294 .4302242
Jabbar .3755303 .277116 1.36 0.176 -.1685103 .9195709
Sade Kala .3485556 .2756082 1.26 0.206 -.1925249 .889636
Sade Khurd -.1911594 .2741581 -0.70 0.486 -.7293931 .3470743
Bankasiya 0 (omitted)

_cons -1.771667 .198216 -8.94 0.000 -2.160809 -1.382524





But when I run areg haz06 odf, absorb(village), I do not get the coefficient on odf:

. areg haz06 odf, absorb(village)
note: odf omitted because of collinearity

Linear regression, absorbing indicators Number of obs = 746
F( 0, 729) = .
Prob > F = .
R-squared = 0.0684
Adj R-squared = 0.0479
Root MSE = 1.2846


haz06 Coef. Std. Err. t P>t [95% Conf. Interval]

odf 0 (omitted)
_cons -1.579263 .0470321 -33.58 0.000 -1.671597 -1.486928

village F(16, 729) = 3.344 0.000 (17 categories)

Can you please explain why that happens?

[Amin Sofla]

I wish you could provide a sample of your data since it will help us to answer your questions easier.
The areg and reg with i.id should give the identical results. However, it is possible that their results differ from each other because their computation procedures are different from each other (please check the econometric books for the details). For instance, I speculate that multicollinearity can be at least among the reasons that their results might differ.
Please see the following simulation as a naive example.

clear all
* Imagine we have 17 villages
set obs 17
set seed 1234
gen village_id = _n
  label var village_id "village specific ID"
* creating time constant village heterogeneity
gen c = rnormal()
  label var c "time constant heterogeneity (village specific)"
* creating panel data (years of observations)  
expand 50
bysort village_id: gen year=_n
  label var year "year of observation"
* creating dummies for the villages  
tabulate village_id, generate(village_d)
* creating the independent variable (explanatory variable)
gen  odf = village_id^2+c
  label var odf "explanatory variable X (with time constant and time varying components)"
* creating the error term  
gen u = 400*rnormal()+3*c
  label var u "error term (correlated with unobservables c)"
* creating the dependent variable (outcome variable)  
gen haz06 = odf + u
  label var y "outcome variable"
* ordering data set  
order  village_id year haz06 odf
drop u c  
* setting the panel
xtset village_id year
* checking the number of observations
ssc install unique
unique village_id
* we have 17 unique villages and
unique year
* we have 50 years of observations
* correlation table
pwcorr odf haz06 village_d1-village_d17, star(0.05)
* Now, we run the regressions:
reg haz06 odf i.village_id
areg haz06 odf, absorb(village_id)
xtreg haz06 odf, fe

As seen in the results, in the first regression, odf is not omitted but in the areg and xtreg, the odf is omitted because of collinearity.

Added: Please also note that in your first regression, you have results for 15 villages and not 16, while you have 17 villages. This might indicate that there is an additional collinearity problem in your data. However, I speculate that this additional problem is not the main reason behind different results for the reg and areg.

[Phil Bromiley]

To add to Amin's helpful comment, you'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data.

The most likely reason you don't get a parameter on odf is that odf does not vary within villages so the dummy for each village explains all the variance that odf might have explained. In panel analysis terms, with fixed effects you can't estimate parameters on variables that do not vary within panels.