Alpha and mean inter item correlation
Locked
 
|
Hi
I am currently using SPSS. I am a novice. I have two set of survey done in Stage 1 of my research and after 3 months the same survey was done in Stage 2. I did an exploratory factor analysis but that was thrown out because I have grouped the variables in a theme which my Supervisor said that cannot be done. EFA can only be done with about 10 to 20 variables and not 4 to 5 variables. Now he told me that I need to do an alpha and mean inter item correlation. I can navigate and manage to do this, alpha and mean inter item correlation. My question is do I put both stages of the survey or only 1 stage of the survey result? There are 69 variables and my participants = 157. The result as shown from the output was participants 314 (this means they took both stages/I fed in both stages) with 69 varibles. Does it matter? whether its 1 stage or 2 stages, will the result varies? Thanks. Flint |
|
|
Hello,
I am trying to run moderated regression, but am not confident if I am doing correct and did 2 different ways. My questions are (1) Could anybody let me know which one is correct or if neither is correct what I
should do? And (2) which result of "Sig." should I report as related to the coefficients?
I am attaching one example. Here, a research question is "how EO affects the relationship between DV (VPTOOL2) and affiliation with philanthropic associations."
Also, EO below is labeled as "centEO"[as this was centered] or "INTER_centEO" as part of a moderating (interaction) term, whereas affiliation with philanthropic associations, as "centAFFIL_PHIL".
Thanks much for your help!
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
Both will test the interaction. The first provides statistics related to the additive model (model 1).
REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA CHANGE ZPP /CRITERIA=PIN(.05) POUT(.10) /NOORIGIN /DEPENDENT
vp2 /METHOD=ENTER
cEo cPhil /METHOD=ENTER eo_phil. OR REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA ZPP /CRITERIA=PIN(.05) POUT(.10) /NOORIGIN /DEPENDENT
vp2 /METHOD=ENTER
cEo cPhil
eo_phil. Checkout: Jaccard, J., & Turrisi, R. (2003) Interaction effects in multiple regression. Newbury Park: Sage. I hope this helps, Steve From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Onishi, Tamaki Hello, I am trying to run moderated regression, but am not confident if I am doing correct and did 2 different ways. My questions are (1) Could anybody let me know which one is correct or if neither is
correct what I should do? And (2) which result of "Sig." should I report as related to the coefficients? I am attaching one example. Here, a research question is "how EO affects the relationship between DV (VPTOOL2) and affiliation with philanthropic associations." Also, EO below is labeled as "centEO"[as this was centered] or "INTER_centEO" as part of a moderating (interaction) term, whereas affiliation with philanthropic associations, as "centAFFIL_PHIL". Thanks much for your help!
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Re: Moderated regression
|
|
In reply to this post by tonishi@iupui.edu
The second approach is the correct approach. When you run an interaction (moderation) model, the individual terms reflect the value of that term when all other terms are equal to 0. In this
case that means centEO is the coefficient when centAFFIL_PHIL is equal to zero and INTER_centEOxcentAFFIL_PHIL (which would be the case when centAFFIL_PHIL is equal to zero, since the product of anything and zero is zero. Remember that the interaction term
(INTER_centEOxcentAFFIL_PHIL) is the modification to the slope values of the individual terms. With continuous terms this all becomes somewhat ambiguous and so the strong suggestion I give to everyone is to plot the interactions.
Matthew J Poes Research Data Specialist Center for Prevention Research and Development University of Illinois 510 Devonshire Dr. Champaign, IL 61820 Phone: 217-265-4576 email: [hidden email] From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Onishi, Tamaki Hello, I am trying to run moderated regression, but am not confident if I am doing correct and did 2 different ways. My questions are (1) Could anybody let me know which one is correct or if neither is
correct what I should do? And (2) which result of "Sig." should I report as related to the coefficients? I am attaching one example. Here, a research question is "how EO affects the relationship between DV (VPTOOL2) and affiliation with philanthropic associations." Also, EO below is labeled as "centEO"[as this was centered] or "INTER_centEO" as part of a moderating (interaction) term, whereas affiliation with philanthropic associations, as "centAFFIL_PHIL". Thanks much for your help!
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Re: Alpha and mean inter item correlation
|
|
In reply to this post by flint
see inserted comments.
> Date: Mon, 22 Oct 2012 21:24:20 -0700 > From: [hidden email] > Subject: Alpha and mean inter item correlation > To: [hidden email] > > Hi > I am currently using SPSS. I am a novice. I have two set of survey done in > Stage 1 of my research and after 3 months the same survey was done in Stage > 2. Pre-post pairs allow you to look at consistency across time, in addition to looking at change. > > I did an exploratory factor analysis but that was thrown out because I have > grouped the variables in a theme which my Supervisor said that cannot be > done. EFA can only be done with about 10 to 20 variables and not 4 to 5 > variables. Well, you can *do* FA with most any N, if you are ready to cope with the consequences. Too many variables make for a non-robust solution -- The rule orf thumb of 10 cases per variable suggests 15 should be good. But if there is strong structure, you might have a good solution with 30. Or a sloppy solution with your 69 vars. I've always looked at new data with factor analysis, just for my own information... to confirm, for instance, that the data *do* have the sort of correlations that go along with valid data entry of a scale on a topic. I don't know why your supervisor says it "can't be done" with 5 vars, unless he is sure that you don't want to merely report that they do or don't cohere. (Try it anyway, and see what you can say about it.) > > Now he told me that I need to do an alpha and mean inter item correlation. I > can navigate and manage to do this, alpha and mean inter item correlation. > My question is do I put both stages of the survey or only 1 stage of the > survey result? There are 69 variables and my participants = 157. > > The result as shown from the output was participants 314 (this means they > took both stages/I fed in both stages) with 69 varibles. Does it matter? > whether its 1 stage or 2 stages, will the result varies? > ... I won't say "never do it", but it is rare to want the same cases entered more than once for anything you will publish. Results are usually similar if you do the two periods separately, even if there was an active intervention. But you did not say what sort of items these are, or if there were special circumstances at either period. Patient scores at "intake" may differ a lot from scores 3 months later. With 69 variables, I have always looked at subscales, either from the result of my factoring or from a-priori logic-- Use published data, or "expert opinion" (whoever is available that you can tout as expert) to pragmatically define some relevant latent variables. -- Rich Ulrich |
|
|
Hi Rich
Thanks. I did a t test to see if there are any changes. I did a FA just to see whether d variables can go together before the t test. Initially I fed all 69 variables and the result was garbage. Didn't show anything coz of the large number of variables. A consultant suggested that I grouped the variables into themes. I did that, so there are about 3-5 variables in a theme. I ran that in FA. I thought it make sense coz the results make sense with the cronbach alpha either .7 or below that. When I showed to my Sup, he said nope cannot b done that way coz i hAve grouped them. So I can't do a EFA. But the grouping was done using face validity. I am so confused. Another question, inter item correlation, ok I do stage 1, I can roughly look at the items tat might b correlated. So do I used these items to do a factor analysis? Is that d reason? Asked my Sup, no answer at all. Feeling frustrated especially for a novice and trudging to make sense as to why I am doing this. It did make sense for factor analysis but trying to make sense abt inter item correlation and after this, what should I do. Sorry still need help Flint |
Re: Alpha and mean inter item correlation
|
|
Yep, all you can see from a FA of 5 items is that, yes, these
seem to go together. Too trivial. Include some others that *don't*. And 3 items is too small for anything that is preplanned. I usually like more than 5 items in a scale. I like it, especially, if my whole 69 items would be reduced to 3-5 subscales, or 10 at the most. Is a factoring of 35 still garbage? 23? Since you are a novice, I will mention: Do these as CFA (iterate on the communalities) and look at the Varimax rotation - it is usually more sensible than the unrotated solution. -- Rich Ulrich > Date: Tue, 23 Oct 2012 14:11:20 -0700 > From: [hidden email] > Subject: Re: Alpha and mean inter item correlation > To: [hidden email] > > Hi Rich > Thanks. > > I did a t test to see if there are any changes. > > I did a FA just to see whether d variables can go together before the t > test. Initially I fed all 69 variables and the result was garbage. Didn't > show anything coz of the large number of variables. A consultant suggested > that I grouped the variables into themes. I did that, so there are about 3-5 > variables in a theme. I ran that in FA. I thought it make sense coz the > results make sense with the cronbach alpha either .7 or below that. > > When I showed to my Sup, he said nope cannot b done that way coz i hAve > grouped them. So I can't do a EFA. But the grouping was done using face > validity. I am so confused. > > Another question, inter item correlation, ok I do stage 1, I can roughly > look at the items tat might b correlated. So do I used these items to do a > factor analysis? Is that d reason? Asked my Sup, no answer at all. Feeling > frustrated especially for a novice and trudging to make sense as to why I am > doing this. > > It did make sense for factor analysis but trying to make sense abt inter > item correlation and after this, what should I do. > > Sorry still need help > ... |
Re: Alpha and mean inter item correlation
|
|
I don’t believe he will be able to do CFA with the standard SPSS package, will he? I thought that was only possible with the AMOS package.
I’m not sure I understand why the initial EFA didn’t work out. Certainly it makes no sense to run an EFA on pre-reduced groups, but running the EFA on all
items with an appropriate rotation (Varimax) should tell you if the items hang together in a reduced number of groups. Was this done? I was a bit confused on that matter. Matthew J Poes Research Data Specialist Center for Prevention Research and Development University of Illinois 510 Devonshire Dr. Champaign, IL 61820 Phone: 217-265-4576 email:
[hidden email] From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Rich Ulrich Yep, all you can see from a FA of 5 items is that, yes, these
> Date: Tue, 23 Oct 2012 14:11:20 -0700 |
Re: Alpha and mean inter item correlation
|
|
I said CFA and spelled out in parentheses the conditions
for "common-factor" analysis, namely, iteration to approximate the communalities. "Confirmatory factors" probably need AMOS, as you suggest. That wasn't on the table. By the way: With those 69 variables, I might try a factoring where I placed a fixed estimate of the communalities on diagonal, such as 0.70, since h^2 estimates the reliability as an r. It might not give a decent solution either, but that is one obvious way to improve the chances for smaller loadings for some variables on some factors. If the items are dichotomous, logic says the fill-in number is lower, 0.55 or even 0.40. I don't remember if I have ever tried that for dichotomies -- For them, I've selected out the variables based on mean-levels, and analyzed subsets, since dichotomies are limited to having their best intercorrelations with other scores of similar (or opposite) skew. Thus, stage 1 gives me a bunch of tiny factors which I score up as simple means; and then I factor analyze that much smaller number of scores. -- Rich Ulrich From: [hidden email] To: [hidden email]; [hidden email] Subject: RE: Alpha and mean inter item correlation Date: Tue, 23 Oct 2012 21:41:17 +0000 I don’t believe he will be able to do CFA with the standard SPSS package, will he? I thought that was only possible with the AMOS package.
I’m not sure I understand why the initial EFA didn’t work out. Certainly it makes no sense to run an EFA on pre-reduced groups, but running the EFA on all items with an appropriate rotation (Varimax) should tell you if the items hang together in a reduced number of groups. Was this done? I was a bit confused on that matter.
Matthew J Poes ... [snip, previous] |
|
|
Hello
Sorry i don't have Amos to do CFA. The statistician thy I have consulted should tell me tht FA cannot be done on a pre reduced group. That's how some ( not all) consultants make $ ( my opinion), I will try the factoring as suggested. Currently trying mean inter item. Hopefully his helps. Flint |
| Free forum by Nabble | Edit this page |