Complete Case Analysis

[Alan Jeddi]

Hi,

I have one DV and 20 IVs.
I conducted a series of univariable regressions and multivariable regressions, using my full data set.

The issue is that some individuals have data available on certain variables, but are missing data on other variables.

I have been told I need to now conduct a Complete Case Analysis, where analysis is run only on individuals who have data available for all variables.

Is there a particular code to do this? I am completely lost as to how to do this.

[Marcos Almeida]

You may perform a regression with all variables, then use - if e(sample), or - gen complete = 1 if e(sample) - and use the data if complete equals to 1.

[Carlo Lazzaro]

Alan:
as an aside to Marcos' helpful comment, please note that if missing data in numeric variables are reported as dots (.), Stata will apply listwise deletion and the observations with missing data in any of the variables (dependent and/or independent) will be ruled out from your regression.
However, the main issue is why some units have missing data and, even more interesting for selecting an approach to deal with them, the mechanism and the pattern underlying their missingness.

[Alan Jeddi]

Hi,

I have one DV and 20 IVs.
I conducted a series of univariable regressions and multivariable regressions, using my full data set.

The issue is that some individuals have data available on certain variables, but are missing data on other variables.

I have been told I need to now conduct a Complete Case Analysis, where analysis is run only on individuals who have data available for all variables.

Is there a particular code to do this? I am completely lost as to how to do this.

Thanks Marcos for that advice.
I basically have 10 variables, a dependent variable called A
and then 9 independent variables, B C D E F G.

The tricky part is that B has only been measured on 10% of people; C on 15% of people; D on 85% of people.

I want to run a linear regression of A on B, adjusting for C D E F G, but I only want to include individuals on whom all of B, C, D, E, F, G have been measured.

I have tried doing the following:
keep A if B= . keep A if C = .

But that doesnt really seem to work.

[Richard Williams]

Listwise deletion (aka complete case analysis) is what regress and most estimation commands do by default. You should not need to do any keep commands beforehand.

It does concern me that one variable is costing you 90% of your cases. Are you sure you need it?

[Carlo Lazzaro]

Alan:
you can use -egen- with the rowmiss()- function to select complete case:

egen flag=rowmiss(Alfa-Omega)*replace the fake names within brackets with the real name of your first and last variable, keeping the hyphen in between
reg A <indpepvar> if flag==0

[Richard Williams]

If you want to limit all of your analyses to only cases that have nonmissing data, you can do something like

reg y x1-x19
gen mysample = e(sample)
reg y x1 x2 x3 if mysample

[Richard Williams]

A discussion of basic and advanced techniques for handling missing data can be found at

https://www3.nd.edu/~rwilliam/stats3/MD01.pdf

https://www3.nd.edu/~rwilliam/stats3/MD02.pdf