I would suggest using repeated measures ANOVA.
A group of 8 volunteers performed a task on two different devices and a particular variable was measured. The question to be answered is if there is a difference between results of the two devices.
For each device, 3 measurements were performed for each person. A person had to perform 6 tasks in total.
This setup could be analyzed by a Two Way ANOVA because there are two factors (time(3) and type of device(2)).
My question is about the order in which the measurements were performed. The assumption for ANOVA is that the order is random. In this case it was randomly chosen for a subject to start with three measurements on device A or to start with three measurements on B. So the three measurements within a device were performed after each other. Is this the correct definition of random order?
If this is not the case I think this problem should be handled with a t-test to compare the means of the two devices. Or is there another test to correct for the repeated factor?
Thank you for your reply.
I would suggest using repeated measures ANOVA.
Thank you for the answer Disvengeance.
Another issue is the exclusion of data points. In some cases 2 of the 3 measurements where absolutely not correct. So the number of data points is not equal for all subjects. Does this affect the quality of the proposed test?
It is best to be cautious when excluding data. How can you be sure that 2 of 3 measurements for a person are incorrect? Couldn't it be possible that the other 1 of the 3 are incorrect? Also, this could just be variation in that measurement.
If you have good reason to suspect that the values are incorrect then you could treat them as missing. Depending on how the data are missing and what software you are using will affect the analysis. Most statistical analysis software have a few methods of performing repeated measures ANOVA, some will perform listwise deletion for missing values (i.e. if a subject has 1 missing value that entire subject is removed from the analysis) and some will not perform listwise deletion. If you have missing values in your data it might be best to find a method that will not perform listwise deletion. Missing data isn't ideal, but it's reality.
Last edited by Disvengeance; 07-14-2014 at 11:15 AM.
Disvengeance, thanks for your advice.
We have reasons to exclude some data points since there were some issues with the measurements. So there are missing data points.
I tried to find out in what way the statistical software (SPSS) handles the missing data. It seems that SPSS uses listwise deletion.
This results in the exclusion of 6 subjects, so only 2 remain. I prefer to do a test whereby more subjects are included.
What about excluding 1 time point so that only 2 of the 3 measurements are required? This will give me 4 subjects.
I feel it's more rigorous to analyze the study the way you had planned, rather than adjust the analysis to favor certain results. At least do so so that you check if the results excluding the first time point are consistent with those of the full analysis.
Tweet |