# Fixed effects regression

#### noetsi

##### Fortran must die
I work with a federal agency that has not provided many details of their analysis (we ask, they don't provide it). When we asked what method they were using for a statistical control they responded 'Fixed effect model?" I understand the difference between random and fixed effects regression, but as best I can tell that is not what they are referencing. Anyone know what statistical method fixed effects models (aka regression apparently) is?

Methods
Fixed-effect model (state fixed effect):

Y = a set of individuals with disability characteristics variables from 1 to X and unemployment rate and industry data in state s year t quarter q plus a state fixed effect plus and error term. The results are generated with national data, we have only state data, they seem to run the results without an intercept and then add this state fixed effect to it. Any suggestions at all of what is occurring here would be appreciated. I have no idea what they are doing.

Their economic effects are particularly confusing, They generate the percent in a given industry (say industrial) in your state by quarter for customers. But if you use that to predict results you would be predicting with a constant.....

https://web.b.ebscohost.com/abstrac...e&authtype=crawler&jrnl=19985010&AN=127504222

Last edited:

#### noetsi

##### Fortran must die
I found this on the same topic today...

ystq = β 0 + Σx1 β Xxstq + αs + ε stq

where ystq is the outcome measure (for instance, employment rate, measured as the
percentage participants who are employed after program exit) in state s in year
t quarter q, xxstq is a set of individuals with disabilities characteristics available
from the state, measuring the extent to which these variables have any effect
on employment rate and unemployment rate in state s year t quarter q, αs is the
state fixed effect, and ε stq is the error term.

I remain confused what form of regression is being done here (what fixed effect regression is) or how you could have two intercepts (one of which is not explained). Ignoring that say I estimated everything from data except the fixed state effect. If I was given the fixed effect (RSA calculates it), could I simply add it to the OLS model I generated?

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Fixed effects regression is usually the type or regression everyone conducts, its void of random effects. Some times in repeated measures people will have time varying effects or time varying confounders meaning the value can change like smoking status, while biological sex wouldnt change across time and it is fixed and only needs to be entered in the model once as a baseline covariate. Were you able to get a copy of the paper? If i have time tomorrow i may take a look at it.

#### noetsi

##### Fortran must die
I found a copy of an article by the individual who refers to fixed effect regression? Can I send it to you email. It is a pdf

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Yeah that is fine given it isn't huge.

#### noetsi

##### Fortran must die
I sent you the methods section that talks about the regression. How they generate the fixed effect for the state is not clear in any documentation provided in case you were wondering.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
What you sent just read as typically regression modeling. They just say fixed for clarity since they are talking about state not states data. I didn't see anything confusing.

#### noetsi

##### Fortran must die
I just never had heard of linear regression called "fixed effects" regression - ever. So I wondered if there was something some method I did not know about.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
It would be like, you wouldn't really use the term frequentist unless you were also talking about Bayesian approaches. So since they are talking about potential geographical states where random effects could come into play I think they are just clarifying that they are just using fixed effects.

#### noetsi

##### Fortran must die
that makes sense. They don't provide a lot of feedback...

#### liweitang

##### New Member
you can search in google with the core words "panel data", or reference to the book "Wooldridge, J.M., 2016. Introductory Econometrics A Modern Approach, 6ed.". the essential of the fixed effect model in the panel data is a regression model.