How to include age as a dummy control in a fixed effect regression?

[Viktor Oeberg]

Dear all,

I have a problem understanding how i should specify the contol variable age in the following regression.

 
 xtreg lnsickdays lnmrtax i.year age inkap, fe

I am measuring the elasticitiy between marginal tax and used sickdays between 2 years and i would like to add age dummy (1 = over 45 year) as a control variable to imitiate earlier work on this field. Although i do not know exactly why, it has come to my understanding that using just age alone, like in the code above, will not be the correct way?

I am not interested in a change in results from people becoming 1 year older between the 2 time periods. I am only interested in presenting the different results in used sickdays for people below/above the age of 45 years.

Should i instead use age as a interaction variable together with the i.year?

Example:

 
 xtreg lnsickdays lnmrtax i.year##age inkap, fe

If i managed to explain my problem clearly enough, please help me understand why i should do it a certain way and please also provide me with some sort of interpretation for the different ways.


Have a nice day

Viktor Öberg

[Carlo Lazzaro]

Viktor:
you may want to try:

 xtreg lnsickdays lnmrtax i.year##i.age_cat_45 inkap, fe

As usual, time-invariant predictors will be wiped out by the -fe- machinery.

[Phil Bromiley]

It depends a lot on what you want to allow to have different parameters. I don't think you said it, but I assume the panel is the individual. Carlo's xtreg will allow different dummies for those above and below 45 for each year. That is, it allows the influence of above/below to vary freely by year. Note also, your analysis requires some within-panel variation in marginal tax rate. If the variation in marginal tax rate is simply reflecting over-time changes in individual income, you may want to control for income - otherwise you will have trouble claiming this is a tax effect not an income effect. An alternative would be simply year and age:
g age_cat_45=(age>44) xtreg lnsickdays lnmrtax i.year i.age_cat_45 inkap, fe However, the parameter on i.age_cat_45 will only reflect individuals who are in one year 44 and the next 45 - all the folks who are over or under 45 both years will be taken care of by the fixed effect. This is the time-invariant issue that Carlo notes. Using age instead imposes the assumption that the influence of age is linear. However, often it is better to use a linear model rather than taking a continuous variable and making it into a dummy. The 45-year-old dummy assumes 20 year-olds and 44 year-olds are the same as are 46 year-olds and 80 year-olds.