Hello Ariana,

I would imagine that many of the "hundreds" of potential DV measurements that you have are correlated. This and the sheer number of measurements suggests several thoughts with respect to your analysis:

1. Running hundreds of anovas will unduly increase the likelihood of type I errors across the aggregate set of tests. You'll have to control for this; an adjustment such as Benjamini-Hochberg would be one straightforward method.

2. If the measures are correlated, then a multivariate approach would likely offer better sensitivity than a slew of univariate tests. However, the vast number of DVs is troublesome on its own (all other things equal, the higher the number of DVs, the lower the statistical power--though the sensitivity difference can still exist--there will be for a given sample size).

3. You might be better served by combining the DV measurements into a smaller number (perhaps via factor analysis).

But, for your question about specifying many cycles of a certain subprogram in spss: Some subprograms allow the specification of multiple variables/cycles (e.g., T-TEST) whereas others don't (e.g., UNIANOVA). GLM does, but only for the repeated measures/multivariate aspect in a single analysis. It looks as if you'd have do set up a DO / REPEAT loop.

Good luck with your work.

Hi Ariana Cahn

Before running ANOVA, make sure your data meet some crucial assumptions:

1. Your dependent variable is measured at the continuous level.

2. Your independet variable consists of at least two categorical, related groups or matched pairs.

Then follow the 13 steps in the link below:

https://statistics.laerd.com/spss-tutorials/one-way-anova-repeated-measures-using-spss-statistics.php

Hello Ariana,

I would imagine that many of the "hundreds" of potential DV measurements that you have are correlated. This and the sheer number of measurements suggests several thoughts with respect to your analysis:

1. Running hundreds of anovas will unduly increase the likelihood of type I errors across the aggregate set of tests. You'll have to control for this; an adjustment such as Benjamini-Hochberg would be one straightforward method.

2. If the measures are correlated, then a multivariate approach would likely offer better sensitivity than a slew of univariate tests. However, the vast number of DVs is troublesome on its own (all other things equal, the higher the number of DVs, the lower the statistical power--though the sensitivity difference can still exist--there will be for a given sample size).

3. You might be better served by combining the DV measurements into a smaller number (perhaps via factor analysis).

But, for your question about specifying many cycles of a certain subprogram in spss: Some subprograms allow the specification of multiple variables/cycles (e.g., T-TEST) whereas others don't (e.g., UNIANOVA). GLM does, but only for the repeated measures/multivariate aspect in a single analysis. It looks as if you'd have do set up a DO / REPEAT loop.

Good luck with your work.

Hello Ariana Cahn. David Morse has given you very good advice, although I'm not sure why you see the need for a DO-REPEAT loop, David. As you say, GLM can take a list of variables, and I think that something like this would generate output that includes both multivariate results and univariate results for each Y-variable:

GLM

y1t1 y1t2 y1t3

y2t1 y2t2 y2t3

y3t1 y3t2 y3t3

etc.

BY Patient

/WSFACTOR=Time 3 Polynomial

/MEASURE=Y1 Y2 Y3 etc.

/METHOD=SSTYPE(3)

/WSDESIGN=Time

/DESIGN=Patient.

If you want the univariate results only, you have to just ignore the multivariate test results, as they are produced automatically with no obvious way to suppress them. (Although as I write this, I bet one could do it via the OMS command. And if not, one could certainly direct the desired parts of the output to a new dataset.)

If missing data is possible, it would arguably be sensible to restructure the data from WIDE to LONG (look up the VARSTOCASES command) and then use the MIXED procedure. In that case, you would likely have to write a macro (look up DEFINE-ENDDEFINE) or some Python code to loop through a list of Y-variables. IIRC, there should be an example of how to do a so-called mixed design ANOVA using the MIXED command in the command syntax reference manual.

Arnab Chatterjee, I'm not sure what you meant exactly when you recommended "checking" the normality of the data. I am always mindful of what George Box said about normality in his famous article, Science and Statistics, which can be read here:

• http://mkweb.bcgsc.ca/pointsofsignificance/img/Boxonmaths.pdf

See Section 2.5.

HTH.

Giovane R Sousa

You say it's a one-way ANOVA, but wouldn't it be a two-way, seeing as I have 2 different groups?

I need time, group and time*group interactions.