Calculating annual % changes in panel data

[Paolo Maldini]

I am running a Diff-in-Diff fixed effects model, where I am evaluating a policy's impact at the neighborhood-level, and I am collapsing the outcome variable at the month-level.

However, I am trying to add controls for annual percentage changes for oil prices, which I suspect influences both my treatment and control observations. However, I have oil prices by year and at the country-level, rather than neighborhood-level.

I tried to create a variable capturing yearly percentage changes for oil prices as follows:
```
foreach v of varlist oil_avg {
gen change_`v' = D.`v'
}
replace change_oil_avg=0 if change_oil_avg==.
```

Here is a data example:
```
. dataex numeric_neighborhood oil_avg change_oil_avg hijri_yr month

----------------------- copy starting from the next line -----------------------
[CODE]
* Example generated by -dataex-. For more info, type help dataex
clear
input long numeric_neighborhood float(oil_avg change_oil_avg hijri_yr month)
1 54.13 0 1438 76
2 79.61 0 1431 1
2 111.26 0 1432 4
2 111.63 0 1433 18
2 108.56 0 1434 34
3 64.3 0 1440 110
3 41.96 -22.340004 1441 111
3 41.96 0 1441 112
3 41.96 0 1441 115
3 41.96 0 1441 116
4 41.96 0 1441 113
4 41.96 0 1441 114
4 41.96 0 1441 117
5 111.63 0 1433 25
6 71.34 0 1439 95
7 108.56 0 1434 30
7 52.32 0 1436 53
7 52.32 0 1436 57
7 52.32 0 1436 62
7 43.64 0 1437 65
7 54.13 0 1438 79
```

I then added the "change_oil_avg" variable to the DiD model as follows:
```
xtreg tobacco_sales i.DiD_implementation i.change_oil_avg i.gender i.education i.month, fe cluster(numeric_neighborhood)
```

However, I keep receiving the following error message
```
change_oil_avg: factor variables may not contain noninteger values
```

Here is a data description of the "change_oil_avg" variable:
```
desc change_oil_avg

Variable Storage Display Value
name type format label Variable label
---------------------------------------------------------------------------------------------------------------------------
change_oil_avg float %9.0g
```

[Jared Greathouse]

This?

[Paolo Maldini]

This?

Thanks, I actually saw and tried that as:
```
sort numeric_neighborhood hijri_yr
bysort numeric_neighborhood:gen pchange=100*(oil_avg[_n]-oil_avg[_n-1])/oil_avg[_n-1]
```
The code works, but I keep receiving the same error message.

[William Lisowski]

In your xtreg command from post #1

xtreg tobacco_sales i.DiD_implementation i.change_oil_avg i.gender i.education i.month, fe cluster(numeric_neighborhood)

change_oil_avg is not a categorical variable so it should not have the i. prefix; that is what the Stata error message was telling you.

Similarly, pchange in post #3 is not a categorical so if you included i.pchange in your model that was the source of the error.

[Paolo Maldini]

In your xtreg command from post #1

xtreg tobacco_sales i.DiD_implementation i.change_oil_avg i.gender i.education i.month, fe cluster(numeric_neighborhood)

change_oil_avg is not a categorical variable so it should not have the i. prefix; that is what the Stata error message was telling you.

Similarly, pchange in post #3 is not a categorical so if you included i.pchange in your model that was the source of the error.

Thanks, I saw one relevant solution to recode the variable here: https://stats.oarc.ucla.edu/stata/fa...s-into-groups/

However, if my goal is to capture the annual variation in oil prices and their potential impact on my outcome variable, does it make sense to recode the variable into a categorical one?

[William Lisowski]

It is unclear whether you understood my advice in post #4.

Where your xtreg command now is, for example,

xtreg tobacco_sales i.DiD_implementation i.change_oil_avg i.gender i.education i.month, fe cluster(numeric_neighborhood)

you could remove the i. prefix on change_oil_avg so your model becomes

xtreg tobacco_sales i.DiD_implementation change_oil_avg i.gender i.education i.month, fe cluster(numeric_neighborhood)

and thereby avoid the error message.

Why would you want to convert change_oil_avg to a categorical variable if there is no need to do so?

[Paolo Maldini]

It is unclear whether you understood my advice in post #4.

Where your xtreg command now is, for example,

xtreg tobacco_sales i.DiD_implementation i.change_oil_avg i.gender i.education i.month, fe cluster(numeric_neighborhood)

you could remove the i. prefix on change_oil_avg so your model becomes

xtreg tobacco_sales i.DiD_implementation change_oil_avg i.gender i.education i.month, fe cluster(numeric_neighborhood)

and thereby avoid the error message.

Why would you want to convert change_oil_avg to a categorical variable if there is no need to do so?

Thanks, indeed I missed that and your solution solved the problem!