Would an Exploratory Factor Analysis be appropriate?

[aquabunny101]

Good Afternoon/Morning,

Following on from my previous post;

 http://spssx-discussion.1045642.n5.nabble.com/Can-Kappa-results-be-displayed-graphically-Confused-Newbie-td5718293.html

I'm hoping that someone might be able to advise me on the best course of action for my data (if there is any :s ). I have conducted a Fleiss Kappa, in which, participants were asked to make decisions about patient attachment styles. For each patient they could choose one of 4 statements (A, B, C or D). My results indicate that there might be confusion/too much similarity between B & D. I wanted to explore this further and was originally told that a confirmatory factor analysis might be in order.  

However, though my own research and recommendation of others it appears that an Exploratory Factor Analysis might be more appropriate but;

1) Is it plausible to conduct EFA with only 4 factors
2) If it is possible, can it be done on categorical data? (All other questions seem to be on a Likert Scales)
2) If not, is there any other method I could use to test this?

As I stated in my previous post, I'm entirely new to statistics and I'm trying desperately hard to up skill myself with something that honestly doesn't come naturally to me (I'm still an Undergrad by the way). Therefore, I really apologise in advance if what I'm writing is too simplistic/muddled etc. Please let me know and I will try my best to clarify.

I have read through the archive but cannot find anything that answers my question.

Thank you in advance!

Kate

[Maguin, Eugene]

Kate,
From your earlier posts, it sounds like you have 13 patients who were rated on a single 4 category item by multiple raters with the number of raters rating each patient varying between some low number and as many as 15.
CFA won't work. What do you want to know from your dataset?

Gene Maguin


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of aquabunny101
Sent: Thursday, February 28, 2013 8:01 AM
To: [hidden email]
Subject: Would an Exploratory Factor Analysis be appropriate?

Good Afternoon/Morning,

Following on from my previous post;


http://spssx-discussion.1045642.n5.nabble.com/Can-Kappa-results-be-displayed-graphically-Confused-Newbie-td5718293.html

I'm hoping that someone might be able to advise me on the best course of action for my data (if there is any :s ). I have conducted a Fleiss Kappa, in which, participants were asked to make decisions about patient attachment styles. For each patient they could choose one of 4 statements (A, B, C or D). My results indicate that there might be confusion/too much similarity between B & D. I wanted to explore this further and was originally told that a confirmatory factor analysis might be in order.

However, though my own research and recommendation of others it appears that an Exploratory Factor Analysis might be more appropriate but;

1) Is it plausible to conduct EFA with only 4 factors
2) If it is possible, can it be done on categorical data? (All other questions seem to be on a Likert Scales)
2) If not, is there any other method I could use to test this?

As I stated in my previous post, I'm entirely new to statistics and I'm trying desperately hard to up skill myself with something that honestly doesn't come naturally to me (I'm still an Undergrad by the way). Therefore, I really apologise in advance if what I'm writing is too simplistic/muddled etc. Please let me know and I will try my best to clarify.

I have read through the archive but cannot find anything that answers my question.

Thank you in advance!

Kate



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Would-an-Exploratory-Factor-Analysis-be-appropriate-tp5718294.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

[aquabunny101]

Hi Eugene,

Thank you so much for your response!

Basically, what I'm attempting to find out is if the categories are the same/too similar and should be condensed into perhaps 2 or 3 factors instead of 4.

As in, I'm trying to explore if it would be better to maybe have participants make judgements based on 2/3 categories instead of the 4 they currently do.


Does that make any sense? I'm sorry for my poor explanations.

kate

[David Marso]

Please suggest why/how you would possibly shoehorn this into either a CFA or EFA!
--

aquabunny101 wrote

Good Afternoon/Morning,

Following on from my previous post;

 http://spssx-discussion.1045642.n5.nabble.com/Can-Kappa-results-be-displayed-graphically-Confused-Newbie-td5718293.html

I'm hoping that someone might be able to advise me on the best course of action for my data (if there is any :s ). I have conducted a Fleiss Kappa, in which, participants were asked to make decisions about patient attachment styles. For each patient they could choose one of 4 statements (A, B, C or D). My results indicate that there might be confusion/too much similarity between B & D. I wanted to explore this further and was originally told that a confirmatory factor analysis might be in order.  

However, though my own research and recommendation of others it appears that an Exploratory Factor Analysis might be more appropriate but;

1) Is it plausible to conduct EFA with only 4 factors
2) If it is possible, can it be done on categorical data? (All other questions seem to be on a Likert Scales)
2) If not, is there any other method I could use to test this?

As I stated in my previous post, I'm entirely new to statistics and I'm trying desperately hard to up skill myself with something that honestly doesn't come naturally to me (I'm still an Undergrad by the way). Therefore, I really apologise in advance if what I'm writing is too simplistic/muddled etc. Please let me know and I will try my best to clarify.

I have read through the archive but cannot find anything that answers my question.

Thank you in advance!

Kate

[aquabunny101]

Hi David,

I'm not suggesting that I should 'shoehorn' it into CFA or EFA. I was asking if it was appropriate to do that or whether there was anything that could be used as a next stage analysis or, as appears to be the case, there is nothing that might be in keeping.

Originally, I was told to use CFA by a supervisor (who I think might be leading me up the garden path slightly), I didn't believe this was correct so started exploring alternatives. Someone kindly mentioned looking at EFA as an alternative to see if it was more fitting and I was seeking help and clarification over this as the literature has me exceedingly confused. If neither are correct then that is fair enough.

Unsure as to whether your message just seems terse or is actually meant to be terse but I'm an undergraduate who is a bit lost with quantitative statistics, I'm desperately trying to up skill myself as I'm the bottom rung of the ladder with statistics, I'm without much help from my supervisor and was looking for clarification from people with much more statistical know how than myself, as to whether it was even feasible. If for no other reason than to say I have tried and exhausted all the avenues I can.

Thank you for your consideration.

Kind Regards

Kate  

[Bruce Weaver]

Was Gene's guess about sample size (n=13) correct?  

aquabunny101 wrote

Hi Eugene,

Thank you so much for your response!

Basically, what I'm attempting to find out is if the categories are the same/too similar and should be condensed into perhaps 2 or 3 factors instead of 4.

As in, I'm trying to explore if it would be better to maybe have participants make judgements based on 2/3 categories instead of the 4 they currently do.


Does that make any sense? I'm sorry for my poor explanations.

kate

[aquabunny101]

Hi Bruce,

Thank you for your reply :)

Basically, I have 5 wards of patients, each ward with 12 patients. Staff member assigned to those wards (n=18 on each ward) were asked to rate each patient against 4 categories with regards to which they felt was more like the patient (type A, type B, type C & type D).

I conducted a Fleiss Kappa (with help from a listserv member) for each ward and ended up with 5 sets of Kappa statistics for inter-rater agreement. The results from each ward seem to indicate that there is poor agreement between classifying patients as Type B and Type D.

One reason I thought that this might be the case was if the Type B and Type D are, in essence, the same thing and I was looking for ways that I might be able to test for this in a meaningful, empirical way. My supervisor suggested a CFA, but, I found though research that it would not be appropriate in the least to use that. Someone else, kindly, suggested that perhaps EFA might be more suitable. However, I cannot work out from the literature I have read if even that would be appropriate as I only have 4 factors of categorical level data. Which I'm not sure meets the parameters needed to do EFA.

I'm seeking clarification as I really don't want to use a statistic if it is actually incorrect, as I wouldn't like to report false/meaningless results. I'm not sure whether there is anything more appropriate to use or if there is nothing to test for this.

Thank you in advance for your consideration. It's, honestly, greatly appreciated as I'm so very, very lost!

Warm wishes

kate

[Bruce Weaver]

It sounds like you are in well over your head here, and I would guess that time to get this done is also limited (I think you mentioned a supervisor, which suggests this is for a student project or thesis).  Therefore, I think it would be advisable to try a much simpler approach to addressing your question about whether B and D perceived as the same.

Do you still have access to the raters?  If you do, how about simply asking them to rate the extent to which they view each pair of categories as being the same?   If the statement was, "These two categories are essentially the same", or something like that, you could use the standard 5-point anchors:  Strongly Agree, Agree, Neutral, Disagree, Strongly Disagree.  If B and D are indeed perceived as the same, and all other pairs as different, you might see something like:

Pair   MeanRating
A-B         4.5
A-C         4.4
A-D         4.3
B-C         4.2
B-D         1.2
C-D         4.3

This would not take long, as each rater has just 6 pairs to rate.

If you don't have access to the original raters, you could use other subjects, provided they have the required knowledge of the categories -- and the more similar they are to the original group of raters, the better.

HTH.

aquabunny101 wrote

Hi Bruce,

Thank you for your reply :)

Basically, I have 5 wards of patients, each ward with 12 patients. Staff member assigned to those wards (n=18 on each ward) were asked to rate each patient against 4 categories with regards to which they felt was more like the patient (type A, type B, type C & type D).

I conducted a Fleiss Kappa (with help from a listserv member) for each ward and ended up with 5 sets of Kappa statistics for inter-rater agreement. The results from each ward seem to indicate that there is poor agreement between classifying patients as Type B and Type D.

One reason I thought that this might be the case was if the Type B and Type D are, in essence, the same thing and I was looking for ways that I might be able to test for this in a meaningful, empirical way. My supervisor suggested a CFA, but, I found though research that it would not be appropriate in the least to use that. Someone else, kindly, suggested that perhaps EFA might be more suitable. However, I cannot work out from the literature I have read if even that would be appropriate as I only have 4 factors of categorical level data. Which I'm not sure meets the parameters needed to do EFA.

I'm seeking clarification as I really don't want to use a statistic if it is actually incorrect, as I wouldn't like to report false/meaningless results. I'm not sure whether there is anything more appropriate to use or if there is nothing to test for this.

Thank you in advance for your consideration. It's, honestly, greatly appreciated as I'm so very, very lost!

Warm wishes

kate

[David Marso]

My 'terse' reply was an effort to make you think more clearly about the problem.
i.e. both CFA and EFA are based upon a correlation matrix or a covariance matrix.
It is not obvious what such a matrix would be based on.
What leads you to believe the categories are being confused by the raters?
Also, IIRC your categories are roughly ordinal in nature.
YOU DELETED YOUR ORIGINAL POST DETAILING THIS INFO!!! **PLEASE DON'T DO THIS**
I thought Fleiss was based on strictly nominal responses (I might be VERY wrong there).
You don't indicate how the data are arranged?  What are the rows?  What are the columns?
What info did you glean from the Fleiss analysis?
Have you considered running your remaining analyses as both uncollapsed and collapsed to see if it makes any difference?
My advice would be to hire a tutor or consultant to help sort this for you (avoid anyone that your 'advisor' might recommend as I fear their counsel would be as dangerous and useless as the "advisor's" ).
Alternatively , table the project for now and take a stats course next semester.

aquabunny101 wrote

Hi David,

I'm not suggesting that I should 'shoehorn' it into CFA or EFA. I was asking if it was appropriate to do that or whether there was anything that could be used as a next stage analysis or, as appears to be the case, there is nothing that might be in keeping.

Originally, I was told to use CFA by a supervisor (who I think might be leading me up the garden path slightly), I didn't believe this was correct so started exploring alternatives. Someone kindly mentioned looking at EFA as an alternative to see if it was more fitting and I was seeking help and clarification over this as the literature has me exceedingly confused. If neither are correct then that is fair enough.

Unsure as to whether your message just seems terse or is actually meant to be terse but I'm an undergraduate who is a bit lost with quantitative statistics, I'm desperately trying to up skill myself as I'm the bottom rung of the ladder with statistics, I'm without much help from my supervisor and was looking for clarification from people with much more statistical know how than myself, as to whether it was even feasible. If for no other reason than to say I have tried and exhausted all the avenues I can.

Thank you for your consideration.

Kind Regards

Kate