Fixed effects

For making suggestions and/or requests for new features you'd like added to EViews.

Moderators: EViews Gareth, EViews Moderator

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Fixed effects

Postby startz » Fri Aug 01, 2014 10:20 am

EViews currently lets you estimate fixed effects in two ways: using @expand and using the fixed effects estimator in a panel dimension.

It would be nice if you could estimate fixed effects without being limited to a panel dimension. For example, I have a panel which is students by grade. I want teacher fixed effects, where each teacher has a number of students. (There are too many teachers for @expand.) It's quite common to want a variety of fixed effects even in an unstructured work file.

It would also be nice if one could get standard errors for the fixed effects.

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Fixed effects

Postby EViews Glenn » Sat Aug 02, 2014 6:41 am

1. Structure as an undated panel with teacher as the cx identifier. Link to a new page if necessary to keep original structure as well.

2. We've always viewed them as a nuisance parameter so we haven't done it, and we've only had a couple of requests over the years. It's not hard for the standard case but a bit tricky to find efficient algorithms for more involved cases. We'll put it on the list to consider.

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Fixed effects

Postby startz » Sat Aug 02, 2014 8:56 am

EViews Glenn wrote:1. Structure as an undated panel with teacher as the cx identifier. Link to a new page if necessary to keep original structure as well.

I'm not following. Each teacher has multiple students. So when I restructure this way I get "Indicies do not uniquely identify observations."

PS Note alert spello on "indices."

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13294
Joined: Tue Sep 16, 2008 5:38 pm

Re: Fixed effects

Postby EViews Gareth » Sat Aug 02, 2014 9:18 am

It should ask you if you want to create new identifier series. answer yes.
Follow us on Twitter @IHSEViews

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Fixed effects

Postby startz » Sat Aug 02, 2014 5:07 pm

Thanks guys. I'm part way there.

I had assumed that the identifier series was exactly the same as the xc identifier. I see that isn't necessarily true.

I'm still stuck on getting the dimensions right. My current page is 4 by 11601. The structure is "grade stdntid"

I want dummies for teachers rather than students. The teacher id is tchid. So I changed the structure to "grade tchid" I said no to remove NA and yes to create CELLID.

The new page is 4 by 1341 by 5276.

I ran a regression that included a lagged dependent variable on the original workfile. (Nothing about teachers or student ids.) The header read

Code: Select all

Sample: 1 46404      
Method: Panel Least Squares   
Periods included: 3      
Cross-sections included: 6568      
Total panel (unbalanced) observations: 13504   


then I ran the same command on the new page and got

Code: Select all

Method: Least Squares
Sample: 1 46404   
Included observations: 21806   


So something I'm not doing something right about the panel structure?

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Fixed effects

Postby EViews Glenn » Sun Aug 03, 2014 7:12 am

If you want the fixed effects by teacher then just structure by teacher alone.

Lags iin a panel structured page do not cross seams between units. Different structures will have different lag missings. You should create lagged series under the appropriate structure, then restructure or copy to new pages as desired. Note that this precomputing of lagged endogenous matters when forecasting.
Last edited by EViews Glenn on Sun Aug 03, 2014 11:26 am, edited 1 time in total.

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Fixed effects

Postby startz » Sun Aug 03, 2014 7:24 am

It sounds so obvious when you say it that way...

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Fixed effects

Postby startz » Mon Aug 04, 2014 9:21 am

Between Gareth and Glenn I am up and running.

Still, I'd like to reiterate my suggestion for the ability to do fixed effects without regard to the workfile structure. Something like an @dummy command analogous to the @expand command. Not doing standard errors would be okay. It would even be okay if it only working with OLS.

Fixed effects is something lots of people need to do and climbing the learning curve for restructuring data isn't always necessary.


Return to “Suggestions and Requests”

Who is online

Users browsing this forum: No registered users and 11 guests