regression analysis for very small sample?

#1
hello,
my study is about "The impact of corporate governance on firm performance".
my sample size is 5 compagnies
i have 6 IV and 1 DV

so my Q is, could i use regerssion analysis with such small sample size?

if no, what is the best alternative to choose?

thanks in advance
 

CowboyBear

Super Moderator
#2
You cannot estimate a regression model with 6 IVs and only N = 5. You literally can't even estimate this model, let alone do so with any reasonable statistical power.

You will need to consider whether you can dramatically increase your sample size, or perhaps change your unit of analysis (e.g., to financial performance by year, for many different years, with year nested in company).
 
#4
thanks for the answer
the sudy examines the performance of the sample companies from 2013 till 2015 ; a period of three years.
what to do, pls any sugestion would give a great help to me.

thank you in advance
 
#5
thank you alot for ur answer

the sample n=5, i examine the V for a period of three years, so there are 15 observations.

please help

and thanks alot
 
#6
Have you looked to see if any of the variables are somewhat redundant (contributing similar information) or if any of the variables are supported by prior theory or literature?

You may also be able to bootstrap all the model parameters to get some idea of what's going on despite the small sample (including the model based statistics like (adjusted) R-squared, the MSE/RMSE).

I see this as a situation where you might need to be creative or be purposeful in the variables you select.
 

CowboyBear

Super Moderator
#7
With the year-by-year data your model is technically estimable, but doing so is (imo) effectively pointless. Your sample size is so small that your estimates will be so imprecise as to be almost uninformative. And you have time series data, but insufficient data to identify any correlated error process (which you would need to). I think you need to go back to the drawing board and figure out what would be a more suitable data source to answer the research questions you have. Sorry - this is not a case where a clever statistical approach is going to help.
 
#8
Sorry - this is not a case where a clever statistical approach is going to help.
Yeah, I guess if the OP can get a new and larger sample, that'd be a much wiser approach than trying to come up with a clever solution that would be bogged down in caveats. I also think that this is probably a case where more data is available or the question can be restructured to make a larger data set more attainable. Good point.