Regression with at least 15 observations?

[amanjot_42]

Hi guys,

I am trying to perform regression analysis on my dataset, but I have one condition that the number of observations should be at least 15 in order to do the regression part. My data looks like:

Cusip  Year  SIC Sales Assets

10000  2000  11  12  15

10000  2001  11  13  18

10000  2002  11  10   20

10000  2003  11  9  25

11000  2000  8   5  25

11000  2001  8   6  28

11000  2002  8   7  29

11000  2003  8   5  28

12000  2003  11  6  30

12000  2004  11  7  25

13000  2002  10  9  10

13000  2003  10  10  15

13000  2004  10  5  20

13000  2005  10  5  25

13000  2006  10  6  20

13000  2007  10  7  20

I want to regress sales against assets by industry and years. So, basically, cross-sectional regressions for each year across industries. But for this, I need to have at least 15 observations for each year regression.

Please suggest me!

Regards

[PaigeMiller]

It's not clear what you are asking. Are you saying you want different regressions for every year in your data, and each year has to have at least 15 observations, and your question is what to do if you don't have 15 observations for each year? Well, obviously, running regressions is not possible under the condition that you have at least 15 observations for each year, you don't have enough data. You have to go out and collect more data, or simply not do the regressions.

I have never heard of such a restriction. The whole point of doing any statistical analysis is to see if, in layman's terms, if the signal is greater than the noise. Signal can be greater than noise with 5 observations, or signal can be less than noise with 500 observations. Why don't you just run the regression and see?

[amanjot_42]

Thanks,
I have panel data of firm-years. So, for each year (for different industries), I want to run a cross sectional analysis with at least 15 firms present in one industry sector (SIC).

[PaigeMiller]

So, is your question: what should you do if you don't have 15 data points for each year/SIC combination?

[amanjot_42]

Skip regression for that,