Which tests to use?????

[Raouf]

Hi Everyone,

I am wondering if anyone can help me with finding the suitable tests to apply in my data analysis. I will explain my data and coding first, then I will explain what I need to figure from this data, so please bear with me.

My data is measurements of duration of one sound, or two sounds together. I recorded 14 participants (7 males and 7 females) producing 4 lists of words, three times each word, just as follows



 The first group is singletons. It has 14 words (7 in onset, at the beginning, and 7 in coda, at the end of words). The second group is onset clusters (two sounds at the beginning of the word). This group has 18 words. The third group is coda clusters (two sounds at the end of the word). This group has 10 words. The last group is geminated stops(one long sound). It has 7 words.

 These words were produced 3 times each by 14 speakers, in three speaking rate (normal, slow and fast) for the time being, let's talk about normal speaking rate only. So:

 

1-Singletons (14x14x3= 588)

2- Onsets (18x14x3=756)

3-Codas (10x14x3=420)

4-Geminated (7x14x3=294)

 

For all of these, I have measures of duration (in millisecond), voicing coded in numbers 1-4 ( which can voiced (1), partially voiced(2), devoiced(3) and voiceless (4)), VOT (voice onset time, in millisecond), males coded as 1, females as 2. I have other variables as tokens, position of sound, and repetition 1,2,or3.  in addition to these, I measured the duration of whatever comes between two sounds in a cluster. I call these epenthetic and intrusive vowels (measured in millisecond).


1- I want to deal with all participants first as one group and test if there is one sound, or two sounds are significantly longer than other sounds or clusters within their group. Then:

2- I want to make some comparisons between males and females to see if one group is significantly different.

3- I want to test the difference between members of the same group (if there is one participant that is significantly different)

Other relations I need to know is:

 

Singletons  in onset vs. singleton in coda

onset clusters vs. coda clusters

male singletons vs. female singletons

males coda vs. female codas

singletons in coda vs. geminated

intrusive in onsets vs. epenthetic in coda

relation between duration and voicing.


P.S. I asked the participants to repeat the words, not to compare their repetitions, but to have more tokens, so i think I don't need within subject repeated measures ANOVA.


I appreciate any help with this.

Thank you in advance,
Raouf

[Maguin, Eugene]

Raouf,
I don't know anything about methods in your research area so there are unfamiliar-to-me terms and experimental designs.
Terms: coda, singletons, onset, geminated.
Here is how I summarize your description. I use <DK> to identify something that is unknown but which needs to be known.
7 males and 7 females pronounced/spoke words from four lists of words. Each word was pronounced/spoken three times each at a slow, normal and fast speaking rate. (Thus, each word was spoken nine times in all.) The first group consisted of 14 words, all <singletons DK> (7 in <onset, at the beginning, DK> and 7 in <coda, at the end of words DK>). The second group consisted of 18 words, all with <onset clusters (two sounds at the beginning of the word) DK>. The third group consisted of 10 words, all with <coda clusters (two sounds at the end of the word) DK>. The last group consisted of 7 words, all with <geminated stops(one long sound) DK>.
The DVs were a) the duration in milliseconds of the target sound(s) <consonants/vowels/syllables DK>, b)
voicing coded as  voiced (1), partially voiced(2), devoiced(3) and voiceless (4), VOT (voice onset time, in millisecond), duration of whatever comes between two sounds in a cluster. I call these epenthetic and intrusive vowels (measured in millisecond).

First of all, you have a quadruplely repeated design: four lists, all spoken by each subject, plausibly equivalent words in a list, words repeated three times and words spoken slow, normal, fast.
Second, you have at least one categorical DV (voicing) that might be regarded as ordinal but might also regarded as nominal.

You want to successfully analyze your data but your subject heading and the kinds of questions asked suggests to me that you are in very deep, rough water and you have only swam in a pool before or not at all. I assume you are at a university somewhere. I assume this may well be a dissertation level project. I think you'd really benefit from finding a statistical consultant/statistician at your university to work with you. It's absolutely not my intention to push you off this listserv. I imagine that you need the kind of help that can only be given by sustained interaction with an expert who knows you.

Respectfully, Gene Maguin






-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Raouf
Sent: Tuesday, January 22, 2013 5:55 PM
To: [hidden email]
Subject: Which tests to use?????

Hi Everyone,

I am wondering if anyone can help me with finding the suitable tests to apply in my data analysis. I will explain my data and coding first, then I will explain what I need to figure from this data, so please bear with me.

My data is measurements of duration of one sound, or two sounds together. I recorded 14 participants (7 males and 7 females) producing 4 lists of words, three times each word, just as follows



 The first group is singletons. It has 14 words (7 in onset, at the beginning, and 7 in coda, at the end of words). The second group is onset clusters (two sounds at the beginning of the word). This group has 18 words.
The third group is coda clusters (two sounds at the end of the word). This group has 10 words. The last group is geminated stops(one long sound). It has 7 words.

 These words were produced 3 times each by 14 speakers, in three speaking rate (normal, slow and fast) for the time being, let's talk about normal speaking rate only. So:



1-Singletons (14x14x3= 588)

2- Onsets (18x14x3=756)

3-Codas (10x14x3=420)

4-Geminated (7x14x3=294)



For all of these, I have measures of duration (in millisecond), voicing coded in numbers 1-4 ( which can voiced (1), partially voiced(2),
devoiced(3) and voiceless (4)), VOT (voice onset time, in millisecond), males coded as 1, females as 2. I have other variables as tokens, position of sound, and repetition 1,2,or3.  in addition to these, I measured the duration of whatever comes between two sounds in a cluster. I call these epenthetic and intrusive vowels (measured in millisecond).


1- I want to deal with all participants first as one group and test if there is one sound, or two sounds are significantly longer than other sounds or clusters within their group. Then:

2- I want to make some comparisons between males and females to see if one group is significantly different.

3- I want to test the difference between members of the same group (if there is one participant that is significantly different)

Other relations I need to know is:



Singletons  in onset vs. singleton in coda

onset clusters vs. coda clusters

male singletons vs. female singletons

males coda vs. female codas

singletons in coda vs. geminated

intrusive in onsets vs. epenthetic in coda

relation between duration and voicing.


P.S. I asked the participants to repeat the words, not to compare their repetitions, but to have more tokens, so i think I don't need within subject repeated measures ANOVA.


I appreciate any help with this.

Thank you in advance,
Raouf




--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Which-tests-to-use-tp5717595.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Rich Ulrich]

I, too, don't know these variables, but I have some comments on
the hypothesis testing.   See these, interspersed below.

> Date: Tue, 22 Jan 2013 14:54:44 -0800

> From: [hidden email]
> Subject: Which tests to use?????
> To: [hidden email]
>
> Hi Everyone,
>
> I am wondering if anyone can help me with finding the suitable tests to
> apply in my data analysis. I will explain my data and coding first, then I
> will explain what I need to figure from this data, so please bear with me.
>
> My data is measurements of duration of one sound, or two sounds together. I
> recorded 14 participants (7 males and 7 females) producing 4 lists of words,
> three times each word, just as follows
>
>
>
> The first group is singletons. It has 14 words (7 in onset, at the
> beginning, and 7 in coda, at the end of words). The second group is onset
> clusters (two sounds at the beginning of the word). This group has 18 words.
> The third group is coda clusters (two sounds at the end of the word). This
> group has 10 words. The last group is geminated stops(one long sound). It
> has  7 words.
>
> These words were produced 3 times each by 14 speakers, in three speaking
> rate (normal, slow and fast) for the time being, let's talk about normal
> speaking rate only. So:
>
>
>
> 1-Singletons (14x14x3= 588)

 [snip, other 3 ...]

For the purposes that I see, you can simplify the dataset by
averaging each of the sets of three. 


>
> For all of these, I have measures of duration (in millisecond), voicing
> coded in numbers 1-4 ( which can voiced (1), partially voiced(2),
> devoiced(3) and voiceless (4)), VOT (voice onset time, in millisecond),

 [ snip, other variables ...]

>
> 1- I want to deal with all participants first as one group and test if there
> is one sound, or two sounds are significantly longer than other sounds or
> clusters within their group. Then:

The above tests are certainly Within-subject tests.  So these are
repeated measures tests or paired-t-tests. 

Do you have prior information that tells you that certain
contrasts are interesting?   (Then, look at those in particular
and leave the rest as "exploratory".)  This is an important point here,
one way or another.  Are you trying to *establish*  something, with
some moderate level of confidence?  or, is this thoroughly an
exploratory effort, where you feel comfortable conducting a dozen
or three dozen tests and presenting the simple, nominal p-value for
every test? 

For a strict procedure of testing --
There is a general method that might apply here, of hierarchical
decomposition.  That goes like this:  The "overall test" is whether
there is a particular difference (in "duration", say) in all the places
that duration exists.  Or, is there an interaction?  -- If you are considering
an "overall 5% experiment-wise error", then you can consult the overall
F for the analysis, or use (say) Bonferroni correction :  Is either the
Main effect or Interaction signification at 2.5%? 

If there is not an interaction, then you do not have an excuse for
claiming that the different sources (or types) of duration are "different"
in followup testing.  This sort of decomposition can conceivable by
carried out for several levels. 

Displaying the actual pattern of means is useful both for checking
consistency, and seeing what to expect for further tests. 



>
> 2- I want to make some comparisons between males and females to see if one
> group is significantly different.

This is between subject testing, N=7 versus N=7 ... so you must
be hoping for rather consistent differences for there to be any effect.

The same sort of hierarchical decomposition described above could apply.

>
> 3- I want to test the difference between members of the same group (if there
> is one participant that is significantly different)

This is pretty unusual, with N=7.  You might look into "detecting outliers".
Any outlier here is probably going to contribute so much wild variance that
other between-subject tests won't have a chance to be nominally significant.

>
> Other relations I need to know is:
>
>
> Singletons in onset vs. singleton in coda
>
 [snip, other things]
>
>
> P.S. I asked the participants to repeat the words, not to compare their
> repetitions, but to have more tokens, so i think I don't need within subject
> repeated measures ANOVA.
...

"... to have more tokens"?  This *sounds*  as if you are wanting
to pretend that your N=14 is N=42 ... which is totally illegitimate.

I had your "P.S." in mind when I wrote my first comment, saying that you
should start by averaging each of these sets of three.  You do have a
gain of precision by collecting the scores in triplicate.  This reduction in
the error size should give more help to the Within-Subject comparisons
you described above (compared to the Between-Subject)  because
it will be a larger fraction of that error variance. 

 Between-Subject variance is apt to be large, for 7 versus 7.  Did you do
a power analysis for this experiment?  (Does the literature include similar
hypotheses explored successfully with Ns this size or smaller?)

--
Rich Ulrich