I've befuddled myself - which test to use?

#1
Hello!

I am in need of help as I've given myself quite the headache. I apologise for the forthcoming barrage of words.

I'm designing a study looking at the effects of hormone cycles on dichotic listening. In particular, looking at voice-onset-time (VOT) and correct responses over two sessions.

It will be a repeated measures design but much more than that I'm a bit stuck, so here goes.

Let's say I have a single cohort of participants taking a dichotic listening test twice (different phases of the menstrual cycle [session 1 and 2, one high hormone levels, one low hormone levels], stimuli-order different in each session) and I capture how many correct responses they have from each ear (LE and RE).

So here, the session (test sat within a particular time frame/phase of the menstrual cycle) is the independent variable as the hormones are the changing factor, and correct responses in LE and RE are what change.

First, I need to compare if the phase of the cycle (session 1 or 2) has an effect on the number of correct responses in both ears (I'm looking, initially, for a change in right-ear-advantage [REA, higher frequency of correct responses in RE]). Am I right in thinking that this would mean I need to treat the correct responses in each ear as separate dependent variables?

So 1 IV of 2 levels and 2 DV:
Mean of correct responses in LE-session1 vs mean LE-session2​
Mean of correct responses in RE-session1 vs mean RE-session2​

If this is the case, which test is best to use on this part? I thought a paired t-test or one-way ANOVA, however, as this is not the main focus of the study, I may perform this test post-hoc. So I'm not certain on which test to use at this phase of analysis.

The main factor is voice-onset-time (VOT). VOT will work as 4 categories short-short (SS, where both ears receive a short onset-time syllable), long-long (LL), short-long (SL, left ear short and right ear long etc), and long-short (LS). I'm expecting that SL and LS will favour correct responses in the ear with the long syllable (so SL will likely yield a correct response in the RE etc). So here, is the independent variable the VOT category and the correct responses in each ear the dependent variables? At least within a single session? But what about across both sessions?

So do I have 2 levels for the independent variable (sessions) and 4 dependent variables (correct responses within each VOT category) to examine? Is the paired t-test still appropriate for this, or a one-way repeated measures ANOVA? Or, am I on the wrong track and, by introducing the VOT categories, am I introducing a second IV, in which case I need to treat the sessions as independent groups and use a factorial ANOVA?

Finally, and I hesitate to ask this...what if I went completely insane and wanted to introduce a second cohort, of women using oral contraceptives, into the study and look at all of the above across both groups? Boom, my brain just exploded out of my ear.

Please help, I want to remain loving statistics for as long as I can.

Yours in perpetual confusion,

Agi
 

hlsmith

Not a robit
#2
Yes there were many word in your post. Curious, we talking about humans here? Second how large of a sample size might you be able get? Didnt read everything but why do you care about individual ears?
 
#3
Yes there were many word in your post. Curious, we talking about humans here? Second how large of a sample size might you be able get? Didnt read everything but why do you care about individual ears?

The responses-by-ear approach will both, allow the corroboration of prior research in demonstrating a change in the right-ear-advantage over menstrual phase (also giving me something to write about), and also allows the measuring of the voice-onset-time part which requires knowing if the correct response was in a particular ear in order to see if long-onset-times have an advantage.

Basically, I'm going to have three of the same test (counterbalanced) from each participant (so 1 IV with 3 levels), which yields 8 dependent variables per test. So a total of 24 for comparison. I have no idea how to analyse so many at once (on a practical level). Multiple ANOVAs with bonferonni correction? Or some mystical super-powered method I don't know about yet...
 

Karabiner

TS Contributor
#4
(I'm looking, initially, for a change in right-ear-advantage [REA, higher frequency of correct responses in RE]).
You could perform a repeated-measures analysis of variance, with 2 within-subject factors "phase of mc" and "ear".
If I understood you correctely, then the interaction between those 2 factors is of particular interest for you.

The main factor is voice-onset-time (VOT).
You can include this as an additional within-subjects factor. In that case you have 2x2x4 conditions.

Finally, and I hesitate to ask this...what if I went completely insane and wanted to introduce a second cohort, of women using oral contraceptives, into the study and look at all of the above across both groups?
You can include a between-subjects factor ("group") in your analysis of variance, making it a "mixed" ANOVA.

Wih kind regards

Karabiner
 
Last edited:
#5
You could perform a repeated-measures analysis of variance, with 2 within-subject factors "phase of mc" and "ear".
If I understood you correctely, then the interaction between those 2 factors is of particular interest for you.


You can include this as an additional within-subjects factor. In that case you have 2x2x4 conditions.


You can include a between-subjects factor ("group") in your analysis of variance, making it a "mixed" ANOVA.

Wih kind regards

Karabiner

Thank you so much!

I don't think I made myself entirely clear however, regarding the VOT (or indeed LE-RE), or I'm missing practical application of these in SPSS.

The VOT itself is indeed only 4 variations, but a correct response to a VOT pair is recorded in either ear. So I would have the number of correct responses to a short-long (SL, or LL/SS/LS) VOT combination where the correct response was from the left-ear and right-ear, or N of long-long in the right-ear etc. meaning I get eight columns of info, per phase of the cycle. I'm unsure of how to practically enter this as conditions in SPSS.

Would I have to simply perform these separately, such as within-subject factor [PHASE]x2 with RE responses from both phases as the variables, then a separate one for [PHASE]x2 with LE responses, and so on for each of the VOT-by-ear responses, or is there a practical way to perform these simultaneously?

Thank you again for your response. It is immeasurably appreciated.

Agi
 

noetsi

Fortran must die
#6
I think you need to block the design (that is use a randomized block design) to address this issue (so you are comparing apples to apples) - unless its actually one of your predictors which is unclear to me. I suggest reviewing this issue in a design of experiment book as its beyond my expertise.

"First, I need to compare if the phase of the cycle (session 1 or 2) has an effect on the number of correct responses in both ears (I'm looking, initially, for a change in right-ear-advantage [REA, higher frequency of correct responses in RE]). Am I right in thinking that this would mean I need to treat the correct responses in each ear as separate dependent variables?"

This sounds to me like a nuisance variable you need to control for, but don't really care about.

https://en.wikipedia.org/wiki/Blocking_(statistics)#Randomized_block_design
 
#7
JUST realised I can create a laterality index variable for LE/RE and each VOT grouping, as well as a number of other variables. I'm down to a handful now :)
 
#8
The VOT itself is indeed only 4 variations, but a correct response to a VOT pair is recorded in either ear. So I would have the number of correct responses to a short-long (SL, or LL/SS/LS) VOT combination where the correct response was from the left-ear and right-ear, or N of long-long in the right-ear etc. meaning I get eight columns of info, per phase of the cycle. I'm unsure of how to practically enter this as conditions in SPSS.
Yes, these are 2 within-subjects factors - "variation" (4 levels) and "ear" (2 levels) - resulting in 4x2=8 columns.
If you add the mc factor, then it's 3 within-subject factors, resulting in 4x2x2=16 columns. Each column is defined
by a particular combination of the levels of the 3 factors.

With kind regards

Karabiner