Trying to determine correct statistical test to use - need assistance

[HyunJee]

I am trying to compare two measures of blood pressure taken on a group of 50 people. Both blood pressure measurements are continious and took place at different points in time.

I am trying to determine if the blood pressure stayed the same, increased or decreased between the two points in time.

An individual I worked with suggested a McNemar test, but to my knowledge this is for 2x2 data. Since my measurement of interested in continuous, this will not work.

I appreciate any help or insight you can provide. Thank you.

[Rick_SAS]

Is there some reason that you don't want to use a paired t-test? http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_ttest_a00000...

[Doc_Duke]

If the difference scores do not look normal, then the Wilcoxon signed rank test (on the difference scores) is another possiblity.  You get to that via PROC UNIVARIATE.

If you need to control for other variables (age, sex, etc), then you can use a regression procedure, though with N=50, your options are more limited.

[Ksharp]

You are right. McNemar is used for frequent(category) data not continuous . (For Cross tabulate)

Since your data is from the same subject(patient). Then you should minus these two variable because they might be correlated(from the same people). then test whether  the mean  of differ equal 0.

You can't directly use proc ttest to do paired-test because these two variables are not independent and might not have Normal Distribution.

Ksharp

[Rick_SAS]

The link I provided shows that PROC TTEST has a PAIRED statement for paired comparisons. See also this example: http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_ttest_a00000...

[Ksharp]

It is new for me .thanks.

But proc ttest is only suited for Normal Data.

If they were not Normal?

As Doc said ,Might should be use non-parameter method (proc npar1way 's Wilconx.)

Or log it to transform it as Normal.

Ksharp

[SteveDenham]

The assumption of normality is greatly over-rated/misunderstood for most linear models, including the paired t-test.  So long as the residuals, not the data themselves, are unimodal and relatively symmetric, the asymptotic p values should be a good guidepost.  For most blood pressure data, this will be the case. 

But suppose you are testing the effect of a drug on lowering BP, and further suppose that the population is a mix of normotensive and hypertensives (which could be further divided by the degree of responsiveness to the drug).  Now, you are probably in the realm of non-normality, and other approaches may be needed.

But for this, it is a classic textbook case for application of a paired t-test.

Steve Denham

[Ksharp · Accepted solution]

Hi. Steve.

I know it is pretty robust. especially for large data.

But I still highly recommend to use proc npar1way + Willcox again, To contrast these two result.

Since t statistic estimator is from Normal Distribution.

Ksharp