Help, please - Confusing results and possibly trouble with variables

[vader90]

Hi to all

I am analyzing data from a research and I am very confused. As I am not good with statistics I am seeking help.

The data is from a research to see what influences the height gain in a particular pathology.

The pre and post therapy heights are expressed both in cm and in SDS (Standard Deviation Scores, calculated through some tables that refer to a normal population).

The height gain was calculated both as Post-therapy height (in cm) - Pre-therapy height (in cm) and as Post-therapy height (in SDS) - Pre-therapy height (in SDS).

The results of the analyses executed are confusing as they differ greatly wen using the height gain expressed in cm as the dependent variable as opposed to using the height gain in SDS as the dependent variable.

Also, contrary to what would be expected, the bivariate correlation between these two variables is non significative and has a very low Pearson score, and the scatter plot correlating these two variables is incredibly dispersed.

After thinking about it for a while I have come up with a few questions:

1) Is the operation used to calculate the SDS height gain the source of the problem? Can I actually subtract a SDS from another?

2) Is SDS the same as z-score?

3) If the answer to 1) is "no, that operation is correct" then does anyone have any idea why I am getting these strange results?

Thank you all in advance

[victorxstc]

The results of the analyses executed are confusing as they differ greatly wen using the height gain expressed in cm as the dependent variable as opposed to using the height gain in SDS as the dependent variable.

You should not compare standard deviations.

Also, contrary to what would be expected, the bivariate correlation between these two variables is non significative and has a very low Pearson score, and the scatter plot correlating these two variables is incredibly dispersed.

Have you again tried to correlate the standard deviations?

The answer for your questions is Yes, No, and Nil, respectively.

[vader90]

Thank you for your answer

You should not compare standard deviations.

Not sure if it makes any difference, but I'm not comparing Standard Deviations, rather Standard Deviation Scores.. I'm comparing where the patients' heights position themselves in the general population, e.g. -1.5SD from the normal mean, +0.7SD from the normal mean.. Are you sure that cannot be done? I've seen it done in some articles that I'm using for reference..

Have you again tried to correlate the standard deviations?

The answer for your questions is Yes, No, and Nil, respectively.

I was trying to correlate the height gain expressed in cm with the same gain expressed in Standard Deviation Score..

[vader90]

Ok just ignore what I just wrote.. I am, yes, comparing SDS and checking how far they are from the mean but.. it's not the same mean. They're different populations, one per year.
Thank you for your help

[victorxstc]

Not sure if it makes any difference, but I'm not comparing Standard Deviations, rather Standard Deviation Scores..

Oh I haven't heard of that SD scores.

I was trying to correlate the height gain expressed in cm with the same gain expressed in Standard Deviation Score

If it was standard deviation, correlating height gain in cm units or in SD units would give you perfect positive correlation, as the increases and decreases happened perfectly simultaneously in both groups. I think the same applies to SDS too. So I don't understand why you haven't found any strong correlation.

Ok just ignore what I just wrote.. I am, yes, comparing SDS and checking how far they are from the mean but.. it's not the same mean. They're different populations, one per year.

I couldn't quite figure out your design, but if I was supposed to do a research on height gain of some individuals over time, I just would use a paired t test (or a non-parametric alternative). The test compares the height of each patient with his height after a year, and runs this across the sample, taking the intraindividual variation into account. But to be honest, I couldn't understand your design and research goal.


edit:
apparently this SDS is the same as z-score as you asked before. So please ignore my comments, as they might not be quite relevant to your case! (but still the t test is a good option)

Thank you for opening up this topic and sorry for my fault.

In a quick search I found these which might be of help to future readers, and hopefully for you:
http://www.spirxpert.com/expressing4.htm
http://www.spirxpert.com/expressing3.htm