0
$\begingroup$

We have 1,000 observations for principal component analysis (PCA). In the following principal component regression(PCR) modeling, I found that only 500 of the 1,000 observations having the outcome we are interested. Can I still use the PCA scores from the 1,000 observations in the following PCR which only includes part of those? Any drawbacks to do this way?

We plan to do PCR on multiple outcomes. All of the 1,000 observations having predictor variables (x1, x2,...x15) available. However, some of them have outcome y1 available, some have outcome y2 available. Only about 100 observations having all outcomes(y1,y2...y5) available. If our main goal is the predictive models from PCR at the end, should I use the 100 observations which have everything available from the beginning?

Thanks.

Improve this question
$\endgroup$
4
  • $\begingroup$ Those other 500 are not part of your population of interest and may have a dramatic effect on the covariance matrix. Your second paragraph does not make sense, however. Do you have a major missing data problem? $\endgroup$ – Dave Jun 25 '20 at 1:21
  • $\begingroup$ Thanks Dave. For second paragraph, I mean all of the 1,000 observations having predictor variables (x1, x2,...x15) available. However, some of them have outcome y1 available, some have outcome y2 available. Only about 100 observations having all outcomes(y1,y2...y5) available. Hope that help. If our main goal is the predictive models from PCR at the end, should I use the 100 observations which have everything available from the beginning? $\endgroup$ – Sailynette Garcia Jun 25 '20 at 1:39
  • 1
    $\begingroup$ My suggestion is to delete this question and post a new question where you frame the question as being about missing data. Missing data present one of the nastier problems in machine learning. $\endgroup$ – Dave Jun 25 '20 at 1:48
  • $\begingroup$ Thanks. But I still think it was just a PCR question. I will add missing data tag. $\endgroup$ – Sailynette Garcia Jun 26 '20 at 4:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.