Pick the value Stata omit in fixed-effects regression

[FitzGerald Blindman]

Hey all, happy new year!

I want to ran a study event regression in the following way:

xtset id years_from_first_birth
xi: xtreg income_in_2018_prices i.years_from_first_birth i.year i.age if gender == 0, robust
xi: xtreg income_in_2018_prices i.years_from_first_birth i.year i.age if gender == 1, robust

Now Stata automatically omitted the variable that was created for years_from_first_birth == 5, but I want it to omit the variable for years_from_first_birth == 0. How can I command Stata to do so?

BTW - if you know a better way to regress what I did please tell me, I'm sure I used the best way.

Thank you!

Fitz

[Clyde Schechter]

First, get rid of the -xi:- prefixes. -xi- is largely obsolete, having been replaced by factor-variable notation. (Read -help fvvarlist- to learn about this.) While it still has some uses with older commands that do not support factor-variable notation, most of those commands have themselves been superseded by newer commands that encompass the same calculations. So don't totally forget that -xi- exists, but think twice or more before actually using it.

xtreg income_in_2018_prices ib0.years_from_first_birth i.year i.age if gender == 0, robust
xtreg income_in_2018_prices ib0.years_from_first_birth i.year i.age if gender == 1, robust

will force Stata to use 0 as the reference level for years from first birth. For year and age, Stata will choose according to its default rule, which will be the lowest value in the estimation sample.

By the way, I would expect year and age to be colinear with each other and also with years from first birth within id. So the within-person effects of these variables on income cannot be estimated from these data. You are getting estimation of between-person effects only. If that is what you want, that is fine. But just remember that you cannot draw any conclusions from this analysis about what happens to an individual's income as they age and time goes by. You can only draw conclusions about the incomes of people who differ in age or are observed in different years.

[FitzGerald Blindman]

First, get rid of the -xi:- prefixes. -xi- is largely obsolete, having been replaced by factor-variable notation. (Read -help fvvarlist- to learn about this.) While it still has some uses with older commands that do not support factor-variable notation, most of those commands have themselves been superseded by newer commands that encompass the same calculations. So don't totally forget that -xi- exists, but think twice or more before actually using it.

xtreg income_in_2018_prices ib0.years_from_first_birth i.year i.age if gender == 0, robust
xtreg income_in_2018_prices ib0.years_from_first_birth i.year i.age if gender == 1, robust

will force Stata to use 0 as the reference level for years from first birth. For year and age, Stata will choose according to its default rule, which will be the lowest value in the estimation sample.

By the way, I would expect year and age to be colinear with each other and also with years from first birth within id. So the within-person effects of these variables on income cannot be estimated from these data. You are getting estimation of between-person effects only. If that is what you want, that is fine. But just remember that you cannot draw any conclusions from this analysis about what happens to an individual's income as they age and time goes by. You can only draw conclusions about the incomes of people who differ in age or are observed in different years.

Hi Clyde, first of all - thank you.

When I tried to run just:

 xtreg income_in_2018_prices i.years_from_first_birth i.year i.age if gender == 0, robust

I got an error message from Stata that told me that it can't handle with negative values and the only way I could fix it was by

xi:

.
I also tried

ib

and it didn't work as well, Stata thought it was part of the variable name.

I'm sorry I don't quate the errors or bring screen prints of them, but I'm not next to the computer that this data is on it.

Thank you again,

Fitz

[FitzGerald Blindman]

Hi Clyde, first of all - thank you.

When I tried to run just:

 xtreg income_in_2018_prices i.years_from_first_birth i.year i.age if gender == 0, robust

I got an error message from Stata that told me that it can't handle with negative values and the only way I could fix it was by

xi:

.
I also tried

ib

and it didn't work as well, Stata thought it was part of the variable name.

I'm sorry I don't quate the errors or bring screen prints of them, but I'm not next to the computer that this data is on it.

Thank you again,

Fitz

I simulated something on my own computer and that's what I've got:

[Carlo Lazzaro]

FitzGerald:
can't you transform -var2- in a two-level categorical variable before=0 and after=1 birth?

[daniel klein]

Without loss of information, recode var2 to positive integers, possibly adding value labels to clarify meaning. The actual numerical values won't matter in your regression context, so they might just as well all be positive.

If the actual values of a (categorical) factor-variable don't matter, why does Stata restrict them to positive values? Well, if negative values were allowed, then -42.var2 could mean two things: the negative of 42.var2 or the indicator for when var2==-42. The restriction to positive values makes the meaning unambiguous.

[Clyde Schechter]

As always, dan klein gives excellent advice clearly explained.

Concerning

I also tried Code:

ib
and it didn't work as well, Stata thought it was part of the variable name.

I can only speculate because you don't show the full command in which you tried using ib, but, let me just emphasize that a period (.) is required between the factor variable operator and the variable name. That is, it must be ib0.years_from_first_birth. Without that period, Stata will, indeed, assume that ib0 is part of the variable name. There are no minor details in code.