Close Home Forum Sign up / Log in

Stats Help

M

I am looking for a stats "guru" who can hopefully give me some advice. As a complete novice as far as stas is concerned I am not sure what the best approach to take to analyse my data.

I have a large number of X-Y graphs (amplitude vs Frequency) that i am looking to investigate the correlation between.
The data sets I am looking to compare are two sets of Y values (using the same Y range) (hope this makes sense). I have seen a similar study that used a Z score for the comparison but from what I have read this works best when comparing large data sets (>100 sets of data). As I have only two sets and am just looking at a point by point correlation between them can some one point me in the right direction please.

I have looked at using Graph pad and excel which both have a correlation function built in but not really knowing what the function is doing (the help file assumes you know which type of analysis you require) I am a bit unsure what the results mean.
Hopefully some one can shed some light on this for me.
Thanks in advance
Max

W

Okay, firstly, I'm not a stats guru. You want to compare the data sets to see if there's any association between them? If the data is interval or ratio, you could use Pearson's correlation co-efficient. If the data is ordinal (I know I'm speaking like a social scientist), then you could use the Spearman Rank correlation co-efficient. I've never used Excel, but you can find step by step instructions on how to do it on the internet.
You'll find the correlation co-efficient has a value of between -1 (perfect negative association) to 0 (absolutely no association) to 1 (perfect positive association). You'll have a p-value which will tell you whether the association is 'significant' or not, bearing in mind causality. Is that what you're after? Is there anyway you can post your analysis results on this thread so that someone can have a gander at it?

M

Hi walminskipeasucker,thanks for the reply, I have attached a small portion of some of my data below. What I am trying to achieve is to come up with some measure of how well the two dtaa ranges compare , on a frequency by frequency basis. Does the Pearson correlation you mention compare the data in corresponding rows, for example compare the two data values for frequency 30.02, then the two for frequency 30.08 and so on. Does the r value then give a correlation result of all the rows.

Sorry if this is a bit unclear but I am not really sure how this works.

I have got some other data sets that I have more samples and have worked out a Z score to give an idea of the correlation, but I gather that I need more than two data sets to use this method.

If anyone can follow what I am trying to achieve and can offer some advice I would be really grateful.

Thanks in advance
Max




Frequency Data 1 Data 2
30.02 38.13 40.19
30.08 37.63 39.3
30.14 37.3 39.31
30.2 39.03 40.8
30.26 37.55 39.77
30.32 37.03 38.82
30.38 37.74 39.03
30.44 37.9 38.51
30.5 37.28 39.17
30.56 37.69 38.62
30.62 38.06 38.52
30.68 36.64 38.26
30.74 37.55 39.08
30.8 36.78 39.44
30.86 37.12 38.75
30.92 36.85 38.42
30.98 38.48 38.96
31.04 38.85 38.95
31.1 37.62 39.48
31.16 37.07 39.45
31.22 38.26 40.01
.
.
.
.
.
.
.
.
200 67.11 68.21

======= Date Modified 03 Sep 2010 21:48:22 =======
I'm a bit confused by this. A z score is just making the data standardised - its not a correlation.

Pearsons r will tell you the correlation of 2 things e.g. "chocolate eating and weight are positively correlated r=.08**, the more you eat chocolate, the more weight you gain" sadly :-(

If you are looking to compare

e.g. case 1 across data set 1 and 2, then you are probably going to need a repeated-measures t-test if I am understanding your data correctly.

M

Hi Sneaks, I will try and explain what the data represents.
The two data sets are a measure of the voltage output of the same item measured using two different sets of test equipment. I am trying to get a way of describing how well the two measurement systems compare. In the ideal world the results should be exactly the same, but obviously the ideal world doesn't exist. So I am hoping to get a value that I can use to compare different sets of data without having to resort to the cumbersome descriptive method I am currently using.
The Z score I calculated was by the looks of it a mistake on my part, as I said I am a complete novice when it comes to stats so I am on a very steep learning curve
Max

======= Date Modified 03 Sep 2010 22:02:44 =======
It sounds like you need a 'repeated measures t-test' (please feel free to contradict me someone).

I can explain it better in psychology terms i.e. people so...

its a test that compares across 2 groups - with the same people in each group - so you compare them across two different conditions. So it should say whether there is a significant difference between the two groups - I guess your hoping for a non-significant result!

I would normally conduct such a test on SPSS, but there are other programmes that will probably do the job.

W

"The two data sets are a measure of the voltage output of the same item measured using two different sets of test equipment. I am trying to get a way of describing how well the two measurement systems compare. In the ideal world the results should be exactly the same, but obviously the ideal world doesn't exist. So I am hoping to get a value that I can use to compare different sets of data without having to resort to the cumbersome descriptive method I am currently using."

I think a Bland and Altman plot may be the way to go with this, since you want to compare how well the two measures compare with one another. Bland JM, Altman DG. (1986). Statistical methods for assessing agreement between two methods of clinical measurement. Lancet, i, 307-310 It's good for more than clinical medicine and is used in chemistry and physics. The paper is extremely well written and I'd recommend at least checking it out to see if it's suitable. I could be wrong but I think it does what you're after. Easy to do as well.

good plan wal! (up)

15941