How to replace the missing value using the correlation between x and y?

Hi,

Let's say I have this data:

T x y

------ ------ ------

1 150 7.2

2 150 7.1

3 210 NaN

4 235 6.1

5 280 5.5

I tried to use correlation on MATLAB but it gives me weired output where all the values are NaNs with size 5x5

Is there a way to replace this missing value using the correlation between x and y?

Answer 1

[Ayush Singh]

Hi Omar,

I understand from your question that you want to have the correlation between x and y but at the same time emit the NaN values.

You could use 'rows' , 'complete' name -pair values to avoid the rows with NaN values.

CorrXY = corr (X , Y, 'rows', 'complete')  % Here X and Y are the two variables for which you are finding the correlation

[Omar Mfarij]

Hi Ayush,

Thanks for answering!

I tried this method before, and it returns everything with NaN. That is why it was weird to me!

What I need is to not ignore the missing value. I need to replace this NaN value after I correlate between x and y. I can simply replace it with mean value and then correlate between x and y.

However, the correlation comes first and then replacing the missing value.

Answer 2

[Jeff Miller]

Omar, if I understand what you are trying to do, I would suggest:

• Form a reduced dataset where you drop all rows with NANs. These rows cannot tell you anything about the relation between X and Y since you don't have Y.
• Using regress or fitlm, fit a linear model to predict Y from X within this reduced dataset.
• Going back to the full dataset, compute the predicted Y for each case with NAN based on its X and the linear model just fit.