Fixed effects regression; missing one year from the year dummy

[Christina Kalaitsidou]

Hello everyone,

I tried to find information regarding the issue I face, however I couldn't and decided to post my question.

I ran a panel fixed effects regression in Stata and included a year dummy (i.year) in my controls. The outcome table demonstrates the years from 2015 to 2019 (where 2019 is omitted). However, my year variable includes data from 2014 to 2019.

Can someone advise why is this happening? Is it because Stata holds 2014 as a reference variable? If yes, then why is 2019 omitted?

Thank you in advance!

Kind regards,
Christina

[Andrew Musau]

Usually, this occurs because the estimation sample has only one or zero observations present in the year 2019. Stata and other software packages implement listwise deletion. There may be other reasons, such as collinearity between the time dummies and other variables. However, to check for the former, you can use:

xtreg... i.year, fe
tab year if e(sample)

[Christina Kalaitsidou]

Hi Andrew,

Thank you for your response. I took your suggestion and executed the code. The results demonstrated the below outcome.
tab year if e(sample)

year | Freq. Percent Cum.
------------+-----------------------------------
2014 | 10,705 14.81 14.81
2015 | 12,082 16.72 31.53
2016 | 12,152 16.82 48.35
2017 | 12,786 17.69 66.05
2018 | 12,617 17.46 83.51
2019 | 11,919 16.49 100.00
------------+-----------------------------------
Total | 72,261 100.00

I believe the reason is due to the collinearity between the time variables.

pwcorr y1 y2 y3 y4 y5 y6, star(0.05)

| y1 y2 y3 y4 y5 y6
-------------+------------------------------------------------------
y1 | 1.0000
y2 | -0.2135* 1.0000
y3 | -0.2087* -0.2081* 1.0000
y4 | -0.2057* -0.2051* -0.2005* 1.0000
y5 | -0.2035* -0.2029* -0.1983* -0.1955* 1.0000
y6 | -0.1960* -0.1954* -0.1910* -0.1883* -0.1862* 1.0000

[Andrew Musau]

If you install reghdfe and absorb the panel identifier and the time variable, you will see which variable(s) is(are) collinear with the time dummies. These will drop out of the estimation.

net install reghdfe, from("https://raw.githubusercontent.com/sergiocorreia/reghdfe/master/src/")
reghdfe ..., absorb(panelvar year)

[Christina Kalaitsidou]

Thank you very much, Andrew!!