[Carlo Lazzaro]

Philippe:
not quite.
You should create one variable only that includes all the years the time dimension of your panel dataset is composed of (put differently: you should group together all your -year- dummies in an unique categorical variable, each level of which relates to a given year).
Then you should -xtset- you dataset:

xtset panelid timevar

The -paneid- tells Stata that you have a panel dataset and that you're going to investigate panel-wise heterogeneity.
Eventually, you can add in the righ-hand side of your regression equation, -i.year- among predictors.

I do recommend you to give a comprehensive reading to -xtset- and -xtreg- entries in Stata .pdf manual before starting your analysis: it might sound time-consuming (and this impression is, at least at the beginning, true) but the long-lasting rewarding is worth the effort (see also William Lisowski 's wise recommendation on this forum about reading and re-reading Stata .pdf manual in order to increase your knowledge of this wonderful software and increase yiour statistical skills through its many worked-out examples).

[Philippe Foultier]

Thanks Carlo, I'll try and squeeze in some time to learn Stata better!

Out of interest, is the previous method of me creating a large number of time variables always wrong, or just only wrong for Stata? I am concurrently doing another analysis in EViews and use all these dummy variables there, with me now becoming a bit uncertain about my general approach.

Cheers for any answer.

Best,

Philippe

[Carlo Lazzaro]

Philippe:
I have a scant smattering of EViews; hence I cannot tell.
As far as Stata is concerned, not only is -fvvarlist- notation more efficient than mysefl in creating categorical variables and interactions, but it also has a tight (and auomatic) relationship with two useful Stata commands, such as -margins- and -marginsplot-.

[Philippe Foultier]

Sorry for the late response Carlo, but thank you. I have followed your advice earlier and learned more about Stata. I have now began using it again and found a new, albeit slight, problem. I'll make a new topic for it though, as it does not relate to this original question.

[Carlo Lazzaro]

Phillppe:
the whole process is hopefully made clearer by the following toy-example:

. set obs 4
number of observations (_N) was 0, now 4

.  g id=1

. g Year_2020=1 in 1
(3 missing values generated)

. g Year_2021=1 in 2
(3 missing values generated)

. g Year_2022=1 in 3
(3 missing values generated)

. g Year_2024=1 in 4
(3 missing values generated)

. foreach var of varlist Year_* {
  2. replace `var'=0 if `var'==.
  3. }
(3 real changes made)
(3 real changes made)
(3 real changes made)
(3 real changes made)
 
. g Cum_Year=_n

. label define Cum_Year 1 2020 2 2021 2 2022 3 2023 4 2024

. label val Cum_Year Cum_Year

. tab Cum_Year

   Cum_Year |      Freq.     Percent        Cum.
------------+-----------------------------------
       2020 |          1       25.00       25.00
       2022 |          1       25.00       50.00
       2023 |          1       25.00       75.00
       2024 |          1       25.00      100.00
------------+-----------------------------------
      Total |          4      100.00


. list id Cum_Year

     +---------------+
     | id   Cum_Year |
     |---------------|
  1. |  1       2020 |
  2. |  1       2022 |
  3. |  1       2023 |
  4. |  1       2024 |
     +---------------+

.

[Philippe Foultier]

Thanks for that Carlo, I will absolutely try to implement this much more clean version of my current cluttering dummies when I have time!