Factor Analysis
Locked
12
12
 
|
Thank you Marcos. for that problem, I hv solve it based on comment from Art :D but I have other problem..from parallel analysis, there are only two factor(raw eigenvalue) more than mean eigenvaleu and prcyntile eigenvaleu. For third factor, raw eigenvalue less than random data eigenvalue although eigenvalue more than 1. From O'corner(2000), his mention that Factors or components are retained as long as the ith eigenvalue from the actual data is greater than the ith eigenvalue from the random data. but from my analysis, there are some variable have greater factor loading for factor 3. so, I am confuse now. 2 or 3 factor? literature review: http://web.ncyu.edu.tw/~fredli/sta/Mao-parallel.pdf Date: Mon, 14 May 2012 16:50:37 -0700 From: [hidden email] To: [hidden email] Subject: RE: Factor Analysis http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=search_page&node=1068821&query=matrix+is+not+positive+definite ----
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/Factor-Analysis-tp5707166p5709641.html
|
|
|
In reply to this post by David Marso
Date: Mon, 14 May 2012 16:50:37 -0700 From: [hidden email] To: [hidden email] Subject: RE: Factor Analysis http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=search_page&node=1068821&query=matrix+is+not+positive+definite ----
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/Factor-Analysis-tp5707166p5709641.html
|
|
In reply to this post by Wana
How did you overcome the singular matrix problem?
The ballpark estimates for the number of factors to retain only narrow the number of solutions you need to interpret. If the third factor has 1 very high loading item that is very clean you might use it as a single item variable if it makes sense. Of course that I would be leery of doing this since it does not do better than a purely random variable. It looks like you would be deciding between 1 and 2 factors. For each of the first 2 factors how many items load at least .5 on a factor and no more than .35 on the other? Remember an item can only be used in the scoring key for one scale. What happens with any items that loaded highly on factor 3? Do the sets of clean loading items make sense as constructs? Once you have the scales what do you intend to do with the scores? E.g., do you want to predict some other variables? Art Kendall Social Research Consultants On 5/15/2012 5:50 AM, Wana wrote: ===================== 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
Art Kendall
Social Research Consultants |
Automatic reply: Factor Analysis
|
|
I am PTOB today Tuesday 15 May returning to the office on Wednesday 16 May. For assistance with
the MITRE Innovation Program please contact the [hidden email] For assistance with Project/Portfolio Pages or Discover, please contact the
[hidden email] For assistance with
other Innovation Zone sites, such as CI&T InZone, please contact [hidden email]. Regards, Mary Lou |
|
|
In reply to this post by Art Kendall
1 2 scale
PS1 .426 .096 1 PS2 .646 .123 1 PS3 .896 .155 1 PS4 .691 .140 1 PS5 .209 .094 0 PS6 .315 .212 0 PS7 .612 .444 0 PS8 .489 .420 0 PS9 -.002 .563 2 PS10 .294 .599 2 PS11 .159 .764 2 PS12 .348 .701 2 PS13 .391 .501 0 It looks like there was no counterbalancing of items, i.e., no mix of positively and negatively loading items. _Do the scales make substantive sense?_ the _preliminary_ scoring would be compute somename1 = mean.4(ps1 ps2 ps3 ps4). compute somename2 = mean.4(ps9 ps10 ps11 ps12 ). How does the column "alpha if item deleted look" in RELIABILITY? You may or may not need to drop some item(s) and redo the computes. What do you mean that you had 400 respondents but ended up with 223? Is 223 an achieved sample from an attempted sample? Or is 223 the number of cases left after listwise deletion? You may or may not be able to gain N if certain items are dropped from a list of variables used in listwise deletion and they do not end up on a scale. (This an example of a benefit from keeping your syntax. You can redraft the analysis very easily.) Art Kendall Social Research Consultants On 5/15/2012 10:31 AM, Najihah Azmi wrote: ===================== 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
Art Kendall
Social Research Consultants |
||||||||||||||||||||||||||||||||||||||||||||
|
|
In reply to this post by Wana
CONTENTS DELETED
The author has deleted this message.
|
|
|
Let's try this again. Ok Bhalla, the data is already collected correct? Will this questionnaire ever be used again? If not, then removing questions is pointless, you already collected them. If you are trying to reduce the number of questions for when you use this survey in the future, then removing redundant questions is possible.
A question to be removed (assuming above statement was satisfied) would go like this. All questions which do not load highly on a specific retained factor would be eliminated from the questionnaire. This isn't a redundant question, its simply an un-useful question in understanding that factor construct. Then you would look at the intercorrelation of variables within a given factor. If two variables have a similarly high factor loading, and they are highly correlated with each other (so you would have retained both of them based on their factor loading value), you may consider removing one of them since it isn't adding anything to the factor's construct. It means that, in the future, you can ask less questions, and get nearly the same amount of information about this construct. I would take a step back and consider the questionnaire design, and purpose of factor analysis in it though. Many people develop questions with the intent of analyzing the individual questions. The reasons why this might be bad are numerous, and not worth discussion here. The argument is then that instead of analyzing the individual questions, it's very likely that they combine in some way to reflect a set of latent constructs. Your 48 questions don't really reflect 48 unique idea's, but rather 3-4 unique idea's. When removing questions, remember, you aren't just throwing them away because they are redundant (that may or may not be true), its because they aren't useful in understanding the 3-4 constructs your instrument measures. What if some of those questions, which load very low on any of the 3-4 constructs, gives you important unique information on its own. You may decide to then keep that one question, but not use it within any construct. This comes back to a ve! ry key point though, all of this only matters if you intend to interpret the questions in terms of these 3-4 latent constructs (meaning you convert them to a scale score, or you perform an SEM that accounts for the latent construct). If you are going to examine the items in isolation, then all of this is for naught, it will not help you. The point of the above then becomes, what is the intent of the final HR survey, from an analysis standpoint? Will inferential statistics be generated? Will this be used to quantify a quality, describe a population, predict an outcome? If it's being used to describe a population or a quality of a population, can that be turned into 3 or 4 dimensions which map onto the factors you are retaining? If this is a one shop deal, you have already collected the data, and this is not for creating a new measure in human resources, then don't worry about redundant variables, there is really no such thing in this particular case. You want to remove redundant variables in something like a multivariate GLM, but that isn't the case in factor analysis. You only want to remove items when you are creating a measure and need to shorten it. 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] -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Deepanshu Bhalla Sent: Thursday, May 17, 2012 3:04 PM To: [hidden email] Subject: Re: Factor Analysis I set up a HR questionnaire that is based on employeee engagement . It consists of 48 questions .5000 people respondend this survey. I wish to cut number of questions (i.e. eliminate redundant questions) from the questionnaire. I run factor analysis taking 2 and 3 factors . My questions is "On what basis i would remove questions from the questionnaire". Is rotational matrix the only way to decide the redundant variables ? If not , what the other basis to eliminate the redundant variables ? -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Factor-Analysis-tp5707166p5711588.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 |
|
|
In reply to this post by Wana
CONTENTS DELETED
The author has deleted this message.
|
|
|
When you look at your 3 or 4 factors, you also want to
a) figure out what they mean based on their items; and b) (probably) look at the reliability of the prospective factor. a) Varimax-rotated solutions are usually pretty good in providing distinct scales. When one item is pretty high on two factors, you have a choice - Use it in one, both, or neither. (I've usually placed it where the correlation and shared meaning were both higher; or dropped it.) b) Procedure Reliability provides the internal-reliability coefficient called Cronbach's alpha. Alpha will depend on two things - the average correlation of the items, and the number of items. "More items" says "more reliability". If you have 20 items in one scale, *perhaps* it can be reduced to 10 or 15 without important loss of reliability - so you might shorten the scale, if it is really important to you to have a short scale. On the other hand, I've never seen anyone in clinical research shorten a scale for that reason. - If the items are distinct, every item provides its own nuance to the total score. But I have seen "highly redundant" items dropped when two items accidentally asked the same question, such that the correlation between them was (say) above 0.90. -- Rich Ulrich Date: Fri, 18 May 2012 04:56:05 +0800 From: [hidden email] Subject: Re: Factor Analysis To: [hidden email]
|
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |