what kind of test should I use for before and after treatment

colstat

What test or distribution should I use for before and after treatment?

Say, I have 30 subjects I test their blood before and after a pill they take a pill, what kind of test should I use?
I want to test mean difference, so, paired t-test? or Wilcoxon signed rank test?
The data looks like this

person before after
#1 100 89
#2 90 75
#3 120 100
#4 132 130
... ... ...
#30 141 120

I was told use binomial with normal approximation, but don't know why. I am so confused, how should I tell? someone please help. Thanks so much.

I like Serena

Hi colstat!

It depends on the type of measurements you take.
Are they nominal, ordinal, or interval?
That is, how do you "test the blood"?

Your test results appear to be of type "interval".
For a blood measurement I would assume the measurements to be normally distributed.

So you would use the paired t-test implying the assumption of a normal distribution.

colstat

thanks! :) So, why am I told to use normal approximation to the binomial. I understand the normal part, but not the binomial part, where that does binomial come from?

Here is another site that actually explains it, but I am still confused
http://www.stat.yale.edu/Courses/1997-98/101/binom.htm

Thanks again!

I like Serena

In the example the measurement is whether you attract the disease or not.
What is counted is the proportion of a group that attracts the disease.
This is binomially distributed.
But the proportion of a group can be approximated with the normal distribution.

In your current problem there's nothing that looks like a success-failure type of measurement so binomial does not come into play.

colstat

I was wondering about this, since the researcher himself said "binomial distribution z statistics with continuity correction."
Also, he said "The P values were calculated by comparing the observed proportion based on 29 patients..."

I am just thinking...what in the world?! why?

The article name is "Single-Site Botulinum Toxin Type A Injection for Elimination of Migraine Trigger Points"

I like Serena

Saying something like that implies a success-failure type of measurement.
You would need to dig up some more information about the measurement involved to be able to say more about it...

SW VandeCarr

Are these mean blood pressures, that is (S - D)/3)+D=MBP where S is systolic and D is diastolic? If so the last two are quite high, but the first two are low for either mean or systolic pressures, particularly 75.

In any case, these are considered continuous variables and MBPs or whatever you have (probably diastolic BP), would be expected to have a normal distribution. The correct way to test effectiveness of an intervention is with a placebo treated control group. You would then compare the mean changes in blood pressure based on the normal assumption for N1 and N2.

If you must compare before and after intervention in the same patients (no control), you should nevertheless use the same test for the before and after data since the variances may differ in the two data sets (unless you are able to use a population variance). In this case you would compare the mean before (N1) and after (N2) measured BP values. This, however, is not good science.

moonman239

Agreed.

BTW, in case you don't know what a controlled experiment is, let me explain. A control group is a group of participants who, in your case, don't take the drug. They just basically live their normal, daily lives.

The experimental group, however, is the group that you're testing the effectiveness of the drug on.

You would calculate the mean blood pressure of the control group and the mean blood pressure of the experimental group.

If the blood pressures of both groups appear to be normally distributed, or you have determined that the Central Limit Theorem applies, you would use the t-distribution. If not, and the distribution is symmetric, you would use the Wilcoxon signed-rank test.

SW VandeCarr

A control group should be as much like the treatment group as possible, except for the treatment itself. This usually is achieved by randomization, sometimes aided by matching on key variables.

There are 30 subjects in the treatment group. With a matched control group, this should be sufficient for the Z test. The OP did list values in ascending order, but I'm not sure why. I don't see why a non-parametric test is indicated without more information.