Restricted number of dimensions in GPA()

[Daan Laméris]

Hi everyone!

I’m very happy that I discovered the GPA() function and I’ve been able to successfully reproduce results from a previous study that used a different software. 

Now I’m trying to use the GPA() function for a different study to analyse data collected through free-choice profiling. Although the results seem valid, it looks like the number of dimensions in the partial configurations is restricted to 19. I believe that normally the dimensions should go to the maximum number of variables (in this case 30), and any unused dimensions return 0’s. 

For example, see the attached screenshot; observer9 used 26 terms, but the partial configuration only goes to 19. Observer10 only used 18 terms, and the partial configuration for dimension 19 returns 0’s. However, I think all of the dimensions should go to 30 as this is the maximum number of terms used by the observers. As a result, the consensus configuration, PANOVA results, and correlations between the initial partial configuration and consensus dimensions are all restricted to 19 dimensions.

 

I’ve been trying to figure out why it restricts to 19 dimensions, but I’m stuck at this point. I don’t think this is something in the script, since with the other dataset that I used to reproduce results from another study the number of dimensions went to the maximum number of terms used by the observers.

 Hope anyone can help me with this!

Many thanks in advance,

Daan

My input is:

gpa.res <- GPA(df, group = c(24,26,30,29,28,14,20,20,26,18,21,22,21,17,20,30,30,21,23,24,30,30,30), scale = TRUE, tolerance=10^-10, nbiteration=200)

 

With df being a matrix with 20 observations and 554 variables.

> str(df, list.len=ncol(df))

tibble [20 x 554] (S3: tbl_df/tbl/data.frame)

[Daan Laméris]

After playing around with different datasets, it looks like this is happening because there are fewer items (rows) than the maximum number of variables of a group. In this case, I only have 20 items that were rated by each group, and the maximum number of variables for a group was 30.  It looks like it takes the number of items - 1 to determine the number of dimensions.

The other dataset I used to see if I could reproduce the results had 38 items and a maximum of 24 variables in a group (the results kept 24 dimensions). If I adapted the dataset to 10 items, the number of dimensions was 9 (again: number of items - 1).

I hope I'm explaining this somewhat clearly. Other studies that I've read and use this method (but with different software) don't have this restriction, so I'm wondering if there is a way to circumvent this, or if the results are reliable and I shouldn't worry :).