st: from reg to xtreg, an interaction term change also in its sign, why?

[Luca Fumarco <luca.fumarco@lnu.se>]

Dear Statalister,

I have a problem with a couple of regressions I am running.


1) I run this model:
reg callback dsec m50 m50_sec
(with dsec and m50 dummy variables)
and I get

	Source	SS       df       MS	Number of obs	=     272
			F(  3,   268)	=    2.59
	Model	1.89291395     3  .630971316	Prob > F	=  0.0529
	Residual	65.1659096   268  .243156379	R-squared	=  0.0282
			Adj R-squared	=  0.0173
	Total	67.0588235   271  .247449533	Root MSE	=  .49311

				
	callback	Coef.   Std. Err.      t    P>t	[95% Conf.	Interval]
				
	1.dsec	-.115942   .0839524    -1.38   0.168	-.2812322	.0493482
	m50	-.143846   .0845766    -1.70   0.090	-.3103651	.0226731
	m50_sec	.0263898   .1196094     0.22   0.826	-.2091038	.2618834
	_cons	.6811594   .0593633    11.47   0.000	.5642816	.7980372
				

then I run the same regression but with xtreg, fe
xtreg callback dsec m50 m50_sec, fe
and I get
. xtreg callback i.dsec m50 i.dsec#i.m50 if many_appl>=0, fe
note: m50 omitted because of collinearity
note: 1.dsec#1.m50 omitted because of collinearity

Fixed-effects (within) regression               Number of obs      =       272
Group variable: sit_id                          Number of groups   =       136

R-sq:  within  = 0.1046                         Obs per group: min =         2
between = 0.0193                                        avg =       2.0
overall = 0.0145                                        max =         2

F(2,134)           =      7.83
corr(u_i, Xb)  = 0.0158                         Prob > F           =    0.0006


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

1.dsec    -.115942   .0368203    -3.15   0.002    -.1887663   -.0431178
m50           0  (omitted)
             
dsec#m50 
0 1    -.0263898    .052459    -0.50   0.616    -.1301445    .0773649
1 1            0  (omitted)
             
_cons     .623295   .0318091    19.59   0.000     .5603821    .6862078

sigma_u   .47076084
sigma_e   .21627055
rho   .82572684   (fraction of variance due to u_i)

F test that all u_i=0:     F(135, 134) =     9.33            Prob > F = 0.0000


why does the interaction term change sign?

2) Moreover, how can I interpret the coefficients of the interaction term in xtreg (the second model)?
 0 1 is simply the coefficient of m50
 1 1 is the proper coefficient of the interaction term (since the values of both terms equal 1), but it is now omitted
 As far as I have understood, in this case, I do not really have an interaction term


Thank you very much in any case!
Cordially


Luca Fumarco

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/