[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: -xtnbreg- vs. -nbreg, cluster()-

From   "Brian Karfunkel" <>
Subject   st: -xtnbreg- vs. -nbreg, cluster()-
Date   Mon, 13 Oct 2008 14:24:40 -0700

Hello all,

I'm trying to figure out how, precisely, -xtnbreg- differs from
-nbreg- (with or without the clustering option).

I have monthly safety data from a company over several years. The
company has many different facilities, and I am trying to look at how
characteristics of the facilities affect safety. In particular, there
was a policy change at a certain time (a change which only affected a
subset of facilities), and I want to see if that policy resulted in
fewer accidents.

The data is panel data organized thusly:

month         | facility  | accidents | policy | hours_worked |
2000m1      |  001      |  0             | old     |  1500             |
2005m12    |  001      |  3             | old     |  1625             |
2000m1      |  001      |  0             | old     |  1800            |
2005m12    |  999      |  2             | new    |  1100            |

Because accidents are count data, and there is evidence of
overdispersion, I am using a negative binomial regression:
- nbreg accidents policy hours_worked, cluster(facility) -

(I don't believe that hours_worked is a strict exposure variable,
since higher-production months may have different characteristics than
lower-production months (e.g. new, greener, workers), which is why I
don't use it as exposure explicitly, but include it as an independent

My question is, what is the difference between doing it as I do above,
vs. setting the data as panel (-xtset facility month-) and using
-xtnbreg- (either fe or re) which has no clustering option?

The facilities are (or, I believe they are) homogeneous, so I'm not
looking to extract the fixed effects between facilities, but the
estimates I get from -xtnbreg- are all lower than from -nbreg-. Is
this evidence that my assumption of homogeneity is incorrect? Or am I
otherwise mispecifying my model?

Thank you for any suggestions/assistance you have.

Brian Karfunkel
Research Fellow
*   For searches and help try:

© Copyright 1996–2023 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index