Hi all,

I have a me-specific question about formatting SPSS output from linear discrimination analysis

I'm measuring brain response (microvolts) using EEG while people play a game that involves pressing a button. The outcome measure of the task is reaction time... a "good" trial is one in which the participant presses the button promptly, and a "bad" trial is one in which they zone out and press the button late. I want to train a classifier to see if individual trials can be classified into "fast," "medium" or "slow" based on the brain response. Since every brain is different, want to run a separate model for each person who plays the game. I then want to compare model performance across people to see if there are certain features (for example: participant age) that are associated with better model performance.

SPSS gives a lot of output when you run LDA, and I'm looking to isolate some specific ones. Is there a way I can ask SPSS to tell me only the percent correctly classified (number shown in the yellow highlight in Screenshot1.png)? This will be helpful as I compare model performance across participants. It would be doubly helpful if there was a way that SPSS could organize those values coherently for me. I've attached a screenshot to illustrate what the ideal output would look like (Screenshot2.png).

Thanks so much for your help!

Best,
KW


------------------------------
Kathryn WT
------------------------------

#SPSSStatistics

Hi. OK, so you need a table consisting of (number correctly classified) / (Total N) for both groups.

It sounds like a job for OMS and Python to me.

In very general terms, you use:
1. The OMS command to select only that table from Discrim.
2. Python to get the data out of it, do the right math, and put it in a table.

If you want me to put a prototype together for you, send me the data and syntax.
marcantr@us.ibm.com


------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

Original Message:
Sent: 10/11/2022 11:17:00 AM
From: Rick Marcantonio
Subject: RE: Linear Discrimination Analysis: Saving Output

Hi. OK, so you need a table consisting of (number correctly classified) / (Total N) for both groups.

It sounds like a job for OMS and Python to me.

In very general terms, you use:
1. The OMS command to select only that table from Discrim.
2. Python to get the data out of it, do the right math, and put it in a table.

If you want me to put a prototype together for you, send me the data and syntax.
marcantr@us.ibm.com


------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

This is an example, using one of the datasets we ship with the product.

dataset close all.

get file '/Applications/IBM SPSS Statistics/Resources/Samples/English/Employee data.sav'.

dataset declare d.

oms /select all /destination viewer=no /tag="_n_".
oms /select tables /if subtypes='Classification Results' /destination format=sav outfile=d /tag="x".

DISCRIMINANT
 /GROUPS=jobcat(1 3) /VARIABLES=salary salbegin jobtime prevexp /STATISTICS=TABLE.

omsend tag="x".

dataset activate d.
select if Var2 = "Count".
execute.

delete variables command_ to var3 total.

omsend tag="_n_".

BEGIN PROGRAM.
import spss
dataCursor=spss.Cursor()
data=dataCursor.fetchall()
dataCursor.close()
d=list(data)

N=0
success=0
c=0
r=0

for row in d:
  c=0
  for cell_data in row:
    N=N+cell_data
    if r == c:
      success=success+cell_data
    c=c+1
  r=r+1

pct=success/N
results=[N,success,pct]

spss.StartProcedure("Results", "RESULTS")
tbl = spss.BasePivotTable("Classification Results", "CR", caption="DISCRIM model")
tbl.SimplePivotTable(rowdim="Case",
  rowlabels=['1'],
  coldim="",
  collabels=['N',"N(Success)","Pct(Success)"],
  cells = results )
spss.EndProcedure()

END PROGRAM.​

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------