Forming panel data fixed effects using sub-sampled data

[Aiyana Cookey]

Hello,

I am very new to Stata and i do not know how to form the right regression. I am testing the effect of corruption on economic growth represented in the time period 1996 to 2016. I will be testing the results by splitting the data into developed and developing countries using a dummy variable, developing=0, developed=1. I will also be representing the corruption variable as a lag.

Before forming my regression, i want to group my data using five year periods.
Through doing this

egen country_id = group(country)

gen period1 = 1 if year >= 1996 & year < 2001
replace period1 = 2 if year >= 2001 & year < 2006
replace period1 = 3 if year >= 2006 & year < 2011
replace period1 = 4 if year >= 2011 & year < 2017

collapse(mean) GDPpercapitagrowth RealGDPpercapita FDI Populationrate Corruptioncontrolindex TradeOpenness PoliticalStability, by (period1 country)

If i am to generate the lag, will i have to do this before grouping the data or after ?

[Carlo Lazzaro]

Aiyana:
welcome to this forum.
If you have annual data, why grouping them in 5-year waves?

[Andrew Musau]

Aiyana:
welcome to this forum.
If you have annual data, why groupong them in 5-year waves?

Sometimes you use low frequency data to test the robustness of the results from annual data.

Before forming my regression, i want to group my data using five year periods.
Through doing this

egen country_id = group(country)

gen period1 = 1 if year >= 1996 & year < 2001
replace period1 = 2 if year >= 2001 & year < 2006
replace period1 = 3 if year >= 2006 & year < 2011
replace period1 = 4 if year >= 2011 & year < 2017

An easier way to do this is to use the -ceiling()- or -floor()- functions.

gen period= ceil((year-1995)/5)

If i am to generate the lag, will i have to do this before grouping the data or after ?

Use the lag operators in the regression. You have 4 observations per country after collapsing the data, so taking lags is really stretching your data. But the following will work.

collapse GDPpercapitagrowth RealGDPpercapita FDI Populationrate Corruptioncontrolindex TradeOpenness PoliticalStability, by (period country)
xtset country period
xtreg yvar xvars L.Corruption, fe

[Carlo Lazzaro]

Andrew:
thanks for pointing this out.
I've never came across something similar in health economics.

[Andrew Musau]

It is common in macroeconomics, see e.g., Table 2 in Acemoglu et al. (2008).

Reference

Acemoglu, Daron, Simon Johnson, James A. Robinson, and Pierre Yared. 2008. "Income and Democracy." American Economic Review, 98 (3): 808-42.

[Carlo Lazzaro]

Thanks again, Andrew.

[Aiyana Cookey]

Hello, thank you for this.

When generating my lags, i seem to notice there is an error saying "no observations". I am not quite sure what to do?

[Andrew Musau]

You lose a cross-section with lagging, assuming a balanced panel. If the panel is unbalanced, you can lose even more cross-sections. Secondly, fixed effects requires a minimum of two cross-sections. Doing the math, you start with 4 cross-sections after collapsing the data. You can afford to lose only the first or the last cross-section as a result of missing values. If missing values are not the issue, make sure that you have the correct xtsettings after collapsing the data. From #3

xtset country period

The previous time variable (year) is no longer valid with the collapsed data.

[Aiyana Cookey]

Yes, i do have some gaps in my data. Would that mean i would have to remove some countries or would there be a different command i could use?

[Andrew Musau]

If there were some observations for some countries, then you would not get a "no observations" error. Therefore, dropping some countries will not help. If you can share your dataset after the collapse, we may be able to verify whether the issue is missing observations. If that turns out to be the case, there is not much that you can do beyond finding alternative data sources.

[Andrew Musau]

Can you copy and paste the result of this instead?

dataex if regexm(country, "^A")

[Aiyana Cookey]

Sensitive data removal.

[Andrew Musau]

Thank you. Your data are fine, it is just that you did not create a consecutive period variable, so the lagging will not be implemented correctly. You need

xtset country_id period, delta(5)

or

egen newperiod= group(period)
xtset country_id newperiod

Notice that my suggestion in #3 will create the consecutive period variable.

[Aiyana Cookey]

It works now, thank you