MHB Analyzing Data: Is There a Significant Difference Between Two Screening Tools?

  • Thread starter Thread starter mermaidrunner
  • Start date Start date
  • Tags Tags
    Data
AI Thread Summary
The discussion focuses on analyzing data from two screening tools used on the same group of children to determine referral rates for further testing. Tool A referred 34 kids while Tool B referred 16, raising the question of whether there is a significant difference in referrals. Participants suggest using a paired t-test, but concerns arise due to the pass/fail nature of the screening tools, leading to a recommendation for a nonparametric approach like McNemar's test. There are challenges with setting up McNemar's test in SPSS, and alternatives like a two-proportion z-test are debated, although its suitability is questioned due to the correlation between samples. The conversation emphasizes the need for an appropriate statistical method to analyze the data effectively.
mermaidrunner
Messages
4
Reaction score
0
I gathered some data on kids who were screened for a disorder using two screening tools. Each child was screened using both tools. I wanted to see if one tool referred more kids for further testing than the other. I need to determine if there is a significant difference between the number of referrals of the two tools. Tool A referred 34 kids and tool B referred 16 kids. Best way to analyze in SPSS?
 
Physics news on Phys.org
If I get you right, for each kid you have two samples, one for each tool. Both numbers are continuous and you want to compare the tools ? If so, paired t test
 
Yankel said:
If I get you right, for each kid you have two samples, one for each tool. Both numbers are continuous and you want to compare the tools ? If so, paired t test

Yes, that is correct. Each kid has a result from each tool. There were 51 total kids tested. My main concern is that these screening tools aren't scored on a numerical scale. It is strictly pass or fail - is that going to have an impact on using the paired t test? Should I be looking at something more nonparametrical?
 
mermaidrunner said:
Yes, that is correct. Each kid has a result from each tool. There were 51 total kids tested. My main concern is that these screening tools aren't scored on a numerical scale. It is strictly pass or fail - is that going to have an impact on using the paired t test? Should I be looking at something more nonparametrical?

Hi mermaidrunner. Welcome to MHB!

That sounds like a two-proportion z-test, pooled for $H_0\colon p_1=p_2$.

See for instance this wiki page.
 
Oh, it ain't numerical...you still have this problem that you get pairs, 2 samples for each kid, meaning you have correlation between these samples.

What about the McNemar's test ?
 
I like Serena said:
Hi mermaidrunner. Welcome to MHB!

That sounds like a two-proportion z-test, pooled for $H_0\colon p_1=p_2$.

See for instance this wiki page.

I had some trouble setting up Mcnemars in spss.. it wasn't looking right . Any suggestions?

I actually went ahead and ran the 2 proportion z-test (in Minitab). Is that going to be effective? Just trying to cover all my bases before submitting.
 
I think that the two proportion Z test is not suitable here, you have correlation within kids.

Here is a youtube clips showing how to do McNemar's test on SPSS.

McNemar Test - SPSS (part 1) - YouTube

Minitab doesn't support it, but you can download a macro from their website which does it.
 
Can anyone spare a few minutes (with SPSS) and look over some analyses I ran related to this? Sorry, I'm out of the stats loop and need these for a project...
 
Back
Top