Panel Data Regression

[Vimal Kumar]

I have micro-level data from 7 countries for 3 years (7 Countries * 3 years * 5000 individuals).
The data is not a panel at the individual level. (That is there is no ID at the individual level)
I want to run panel data regression on this data to verify the country effect and year effect.

I am new to panel data regression.
Can someone suggest me how to go about this?

[Carlo Lazzaro]

Vimal:
welcome to this forum.
As per your description, in your case countries represent your -panelid-.
By the way, there's no methodological reason that supports individual as the only possible -panelid-.
That said, if you have data on individuals nested within countries (or other higher level variable that expresses grouping), you should consider -mixed-.

[Vimal Kumar]

Sir, Thank you for quick your response.

I tried running panel regression. Stata says the ID and Year combination should be unique.
But in my case, as each country has individual data, I will have many Year * Country combination.
For example, Kenya has n = 3000. So for 3 years, we have 3 * 3000 = 9000 records. Then similarly for Tanzania or India.
But, Stata expects, country * year to be unique. India_ 2015 one record, India_2016 one record and so on.
So, now I am confused what do I do with 3000 individuals. What should be my ID?

Objective my study is to compare one country with the other and within-country how things have change year by year.

[Carlo Lazzaro]

Vimal:
if you do not plan to use time-series commands, such as lags and leads, you can simply:

xtset <panelid>

without a <timevar>.

[Vimal Kumar]

Sir,

Here is a screenshot of my dataset.
The country variable has seven different values (India, Bangladesh, Kenya and so on) and Year Variable has 3 years (2015, 2016, 2017)

When I try xtset:

. encode country, gen(cntry)

. xtset cntry year
repeated time values within panel
r(451);

I am getting the error, "repeated time value within panel"...

Please suggest me some workarounds...

Thank you,

Regards,
Vimal

[Carlo Lazzaro]

Vimal:
- as per FAQ, please do not post screenshots, but share what you typed and what Stata gave yopu back via CODE delimiters and/or provide and excerpt/example of your data via -dataex-. Thanks;
- as per my previous reply, you can avoid Stata error message via:

xtset country

The abovementioned workaround implies that you do not plan to use -tsvarlist- commands, such as leads and lags, in your panel data regression.

[Vimal Kumar]

Hello Sir,

Sorry for the screenshot. Will not post again.

I ran the code as you mentioned:

. xtset cntry
       panel variable:  cntry (unbalanced)

. xtologit useph age ppiscore

Fitting comparison model:

Iteration 0:   log likelihood = -211781.22  
Iteration 1:   log likelihood = -205756.18  
Iteration 2:   log likelihood = -205697.47  
Iteration 3:   log likelihood = -205697.44  
Iteration 4:   log likelihood = -205697.44  

Refining starting values:

Grid node 0:   log likelihood = -205680.04

Fitting full model:

initial values not feasible
r(1400);

I am getting the error: "Initial Values not feasible".

Can you help me with this?

Thank you again,

Regards,
Vimal


.

[Carlo Lazzaro]

Vimal:
see Technical note on quadrature, -xtologit- entry, Stata .pdf manual, page 312.

[Vimal Kumar]

Dear Sir,

Thank you for your patience.
I am sorry. I am really new to this.

I tried changing intpoints(), as mentioned for quadrature.

I am getting the same error.

Thank you,

Regards,
Vimal

[Carlo Lazzaro]

Vimal:
see -quadchk- entry in Stata .pdf manual.
Otherwise, check whether your regression model is correctly specified.