Factor analysis
Locked
 
|
Hi all, I am struggling with factor analysis that is new to me. I have a 20-item likert questionnaire and I ran an initial solutions and the component analysis table below I cannot understand. Some items which have E, would do they mean? Would anyone help? Thanks.
Component Matrix
Extraction Method: Principal Component Analysis. a 5 components extracted. |
|
|
Hi Jacqueline, "E" is a common way how to write numbers in
the Scientific notation (http://en.wikipedia.org/wiki/Scientific_notation).
For example 3.189E-02 means 3.189 * 10 ^ -2 = 0.03189.
In the factor analysis, you can usually ignore these
numbers as if they were zeroes - they are too small to have any meaning. You can
drop them from the output by using
/FORMAT BLANK(.10)
option in the syntax
(or check the "Suppress absolute values less than..." option in the
interface).
Best,
Jan
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of jacqueline london Sent: Friday, September 03, 2010 10:17 AM To: [hidden email] Subject: Factor analysis Hi all, I am struggling with factor analysis that is new to me. I have
a 20-item likert questionnaire and I ran an initial solutions and the
component analysis table below I cannot understand. Some items which have E,
would do they mean? Would anyone help? Thanks.
Component Matrix
Extraction Method: Principal Component Analysis. a 5 components extracted. _____________ Tato zpráva
a všechny připojené
soubory jsou důvěrné a určené výlučně adresátovi(-ům). Jestliže nejste
oprávněným adresátem, je zakázáno jakékoliv zveřejňování, zprostředkování nebo
jiné použití těchto informací. Jestliže jste tento mail dostali neoprávněně,
prosím, uvědomte odesilatele a smažte zprávu i přiložené soubory. Odesilatel
nezodpovídá za jakékoliv chyby nebo opomenutí způsobené tímto
přenosem.
P Are you sure that you
really need a print version of this message and/or its attachments? Think about
nature.
|
Administrator
|
In reply to this post by jacqueline london
Jan already addressed your question about the meaning of the E. I'm writing to comment on your choice of PCA as the extraction method. Often, people use it because it is the default setting. But in many cases, it would be more appropriate to use some form of factor analysis per se. For more on the distinction, see the "EFA Versus PCA" section of Preacher & McCallum's nice article.
http://www.people.ku.edu/~preacher/pubs/preacher_maccallum_2003.pdf HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
In reply to this post by jacqueline london
Do Likert data warrant this approach? Are your items replications of
previous work by yourself or others, or are they newly developed (for a PhD thesis)? You can usually get what you want visually from an initial correlation matrix, preferably non-parametric, but we all do it, and if it's for a PhD I suppose you have to go throught the motions for the assessors. John Hall [hidden email] http://surveyresearch.weebly.com ----- Original Message ----- From: jacqueline london To: [hidden email] Sent: Friday, September 03, 2010 10:16 AM Subject: Factor analysis Hi all, I am struggling with factor analysis that is new to me. I have a 20-item likert questionnaire and I ran an initial solutions and the component analysis table below I cannot understand. Some items which have E, would do they mean? Would anyone help? Thanks. Component Matrix Component 12345 SELBEL1.399.252-.398-.280.105 SELBEF2.251-.564.263.167.331 SELBEF3.647-.224.129-.1396.780E-02 SELBEF4.627-.146.190-.2632.454E-02 SELBEF5.331.469.1148.683E-02.279 SELBEF6.460.215.423-4.469E-02-.127 SELBEF7.362-.1593.189E-02-.382-.148 SELBEF8.634-1.308E-025.161E-02-1.292E-02-.321 SELBEF9.405-1.846E-02.135-.321-.369 SELBEF10.497-.4646.585E-026.435E-02.118 SELBEF11.274-1.904E-02-.250.713-.242 SELBEF12.576.250-.264-1.427E-02-.282 SELBEF13.350.438.356.142.269 SELBEF14.395.214.475.287-.192 SELBEF15.464.3645.913E-02.153.374 SELBEF16.604-.3731.975E-02-1.295E-02-1.068E-02 SELBEF17.488.338-.319-.278.334 SELBEF18.378-.402-.138.154.217 SELBEF19.539.220-.194.214-.270 SELBEF20.658-.189-.444.166.113 Extraction Method: Principal Component Analysis. a 5 components extracted. ===================== 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 Spousta Jan
|
|
|
The critic was probably asking for something like Cramer's phi
or V.
See the Wikipedia entry:
Given that your situation involves two ordinal variables
instead of
two nominal variables, there may be an effect size measure
other
than Cramer's phi/V which uses the the relative magnitude
information.
At first glance, gamma seems to be such a measure but a quick
search of the literature does not turn up any instances of
gamma
being used as an effect size measure; for the gamma test
see:
I assume that someone with more experience with contingency
table
analysis might be able to provide more clarity on the
issues.
-Mike Palij
New York University
|
|
|
Hi:
What about Kendall's tau-b? HTH, Mart GG Mike Palij wrote: > The critic was probably asking for something like Cramer's phi or V. > See the Wikipedia entry: > http://en.wikipedia.org/wiki/Effect_size#.CF.86.2C_Cram.C3.A9r.27s_.CF.86.2C_or_Cram.C3.A9r.27s_V > > Given that your situation involves two ordinal variables instead of > two nominal variables, there may be an effect size measure other > than Cramer's phi/V which uses the the relative magnitude information. > At first glance, gamma seems to be such a measure but a quick > search of the literature does not turn up any instances of gamma > being used as an effect size measure; for the gamma test see: > http://en.wikipedia.org/wiki/Gamma_test_%28statistics%29 > > I assume that someone with more experience with contingency table > analysis might be able to provide more clarity on the issues. > > -Mike Palij > New York University > [hidden email] <mailto:[hidden email]> > > > > ----- Original Message ----- > *From:* Eins Bernardo <mailto:[hidden email]> > *To:* [hidden email] <mailto:[hidden email]> > *Sent:* Saturday, September 04, 2010 11:31 AM > *Subject:* Effect size > > Dear all, > > My apology for cross-posting .... > > A friend was using a chi-square test to test the null hypothesis > that the two ordinal variables,each with three catergories, are > not associated. The frequency counts on the cross-tab, chi-square, > df, p-values as well as the gamma coefficient are reported. One > critic was asking to include the effect size. Is gamma > coefficient not the effect size? > > Thank you in advance for your comments. > > Eins > > -- For miscellaneous SPSS related statistical stuff, visit: http://gjyp.nl/marta/ ===================== 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 |
|
|
----- Original Message -----
On Sunday, September 05, 2010 4:39 AM, Marta García-Granero wrote: > Hi: > What about Kendall's tau-b? Before responding to Marta's comment, I'd like to make a few general points: (1) A data analyst needs to take responsibility for the analysis he/she is conducting which means (a) an analysis should not be undertaken unless the analyst is thoroughly familiar with the statistical analysis (in contrast to being familiar with the syntax to perform an analysis with some canned stat package) and (b) knowing what the conventions are that are being used both in one's specific area (which may be biased towards the reporting of specific statistics even when doing so might be questionable) and more general practice. Sometimes (b) takes precendence over (a), especially if reviewers rely on certain heuristics (i.e., use test X in situation A) in contrast to actually understanding why the statistical analysis is appropriate for the situation being presented. In these cases, one will have to know what is the generally accepted heuristic that is being followed. (2) When one asks for advice on a particular statistical issue one must be aware that there may not be a single best solution or single agreed on solution. If one is sufficiently familiar with the fundamentals of a statistical analysis that was done, one can argue logically why certain things were done, why certain decisions were made, and why certain statistics are being reported. This is a statistical criterion. However, depending upon a "critic's" or a reviewer's level of sophistication (which can vary tremendously), such statistical justifications may not matter, in part, because the critic/reviewer may not accept or understand it. In this case, "best statistical practices" may be less important than "commonly used statistical practices". The former requires an understanding of the statistical literature, the latter requires an understanding of what the "audience" is willing to accept in one's presentation. (3) For the problem at hand, it would be useful to examine some references to get an idea of what might be appropriate. In the case presented below, one might want to take a look at Robert Grisson & John Kim's "Effect Sizes for Research: A Broad Practical Approach" which is available in partial preview mode (i.e., chunks of the book cannot be read) on books.google.com (search for the authors; the URL is too long to present here). One might want to skim the whole book to understand the framework being used but Chapter 9 "Effect sizes for ordinal categorical variables" would be most relevant. However, the real usefulness of this chapter and this book is being able to follow up on the references provided in order to determine what is the best thing to do in one's situation. (4) Contingency table analysis is not my thing, so though I have some sense of the basic statistical issues and the sociology of statistical use issues, I don't claim expertese in this area. I presume that some people on this list may have the appropriate expertese or one might be able to locate such people on other lists or by referring to the statistical research literature. The point is that I may be able to recommend the use of Cramer's phi/V because that seems to be a popular choice (in part, I think, because it can be used in power analysis), I cannot explain why other measures that seem to be reasonable alternatives are not used. For example, Marta above suggeted the use of Kendall's tau-b. Though Thomas Wickens (1989) in "Multiway Contingency Tables Analysis for the Social Sciences" talks about effect sizes in this type of analysis (see his chapter 9 "Measures of effect sizes") he does not include Kendall's tau (a and b) as an effect size measure, rather he covers Kendall's tau (a and b) as measures of concordance in chapter 13 on "Ordered categories". Similarly, Grisson and Kim do not use Kendall's tau-b as an effect size measure. That being said, an article by Berry et al (2009) explicitly label Kendall's tau (both a and b) as effect size measures though they argue for the use of Stuart's tau (which they call tau-c) as a more appropriate statistic. The abstract for this article is available on PubMed; see: http://www.ncbi.nlm.nih.gov/pubmed/19897822 The article appears in the journal "Behavior Research Methods", a journal of the Psychonomics Society and should be avaialable at most university libraries. One conclusion that one might arrive at here is that there is more going on than is being explicitly stated. (5) So, which effect size measure should one use in this situation? If this is a topic that is near and dear to one's heart, one will have do a fair amount of background reading to find out what has been done, what is currently being recommended, and what problems exist in this area. If one just wants to shut a critic/reviewer up, then one might ask that critic/reviewer what effect size statistic would they find acceptable (don't be surprised if their response makes no sense) and provide that. The simplest solution would be to ask what is the most commonly used effect size measure in this situation. I've provided one possibility but that might not be the best (though acceptable) answer. -Mike Palij New York University [hidden email] > Mike Palij wrote: >> The critic was probably asking for something like Cramer's phi or V. >> See the Wikipedia entry: >> http://en.wikipedia.org/wiki/Effect_size#.CF.86.2C_Cram.C3.A9r.27s_.CF.86.2C_or_Cram.C3.A9r.27s_V >> >> Given that your situation involves two ordinal variables instead of >> two nominal variables, there may be an effect size measure other >> than Cramer's phi/V which uses the the relative magnitude information. >> At first glance, gamma seems to be such a measure but a quick >> search of the literature does not turn up any instances of gamma >> being used as an effect size measure; for the gamma test see: >> http://en.wikipedia.org/wiki/Gamma_test_%28statistics%29 >> >> I assume that someone with more experience with contingency table >> analysis might be able to provide more clarity on the issues. >> >> -Mike Palij >> New York University >> [hidden email] <mailto:[hidden email]> >> >> >> >> ----- Original Message ----- >> *From:* Eins Bernardo <mailto:[hidden email]> >> *To:* [hidden email] <mailto:[hidden email]> >> *Sent:* Saturday, September 04, 2010 11:31 AM >> *Subject:* Effect size >> >> Dear all, >> >> My apology for cross-posting .... >> >> A friend was using a chi-square test to test the null hypothesis >> that the two ordinal variables,each with three catergories, are >> not associated. The frequency counts on the cross-tab, chi-square, >> df, p-values as well as the gamma coefficient are reported. One >> critic was asking to include the effect size. Is gamma >> coefficient not the effect size? >> >> Thank you in advance for your comments. >> >> Eins >> >> > > > -- > For miscellaneous SPSS related statistical stuff, visit: > http://gjyp.nl/marta/ > > ===================== > 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 E. Bernardo
Hi,
Can you please help?
I’d like to add a loop to the following syntax that will allow me to change the reference category in the syntax below. I have a variable
called ‘column’ which is my independent variable and a variable called binaryrow, which is my dependent variable. I’d like to change the reference category of my independent variable ‘column’ so that I loop through all possible categories of the column variable.
E.g. if my column variable has 9 categories then I would like to run the logistic regression 9 times, with the reference category being 1 in the first model, 2 in the second, 3 in the third and so on.
WEIGHT OFF . CSPLAN ANALYSIS /PLAN FILE='C:\blahblahlah /PLANVARS ANALYSISWEIGHT=svyweight /SRSESTIMATOR TYPE=WR /DESIGN strata= svystrata CLUSTER= svypsu /ESTIMATOR TYPE=WR. CSLOGISTIC binaryrow (HIGH) BY column /PLAN FILE='C:\plan.sav' /MODEL column /INTERCEPT INCLUDE=YES SHOW=YES /STATISTICS PARAMETER EXP
/TEST TYPE=CHISQUARE PADJUST=LSD /ODDSRATIOS FACTOR=[column(9)]
/MISSING CLASSMISSING=EXCLUDE /CRITERIA MXITER=100 MXSTEP=5 PCONVERGE=[1E-006 RELATIVE] LCONVERGE=[0] CHKSEP=20 CILEVEL=95 /PRINT SUMMARY VARIABLEINFO SAMPLEINFO HISTORY(1). Many Thanks Jamie
================================================================================================ This e-mail and all attachments it may contain is confidential and intended solely for the use of the individual to whom it is addressed. Any views or opinions presented are solely those of the author and do not necessarily represent those of Ipsos MORI and its associated companies. If you are not the intended recipient, be advised that you have received this e-mail in error and that any use, dissemination, printing, forwarding or copying of this e-mail is strictly prohibited. Please contact the sender if you have received this e-mail in error. Market & Opinion Research International Ltd (trading as Ipsos MORI), Registered in England and Wales No. 948470, 79-81 Borough Road, London SE1 1FY, United Kingdom. VAT reg no. 443932151. Email: [hidden email] ================================================================================================ |
|
Administrator
|
You could stick your CSLOGISTIC command in a macro, like this: define !myloop (N = !tokens(1)). !do !i = 1 !to !N CSLOGISTIC binaryrow (HIGH) BY column /PLAN FILE='C:\plan.sav' /MODEL column /INTERCEPT INCLUDE=YES SHOW=YES /STATISTICS PARAMETER EXP SE CINTERVAL TTEST DEFF DEFFSQRT /TEST TYPE=CHISQUARE PADJUST=LSD /ODDSRATIOS FACTOR=[column(!i)] /MISSING CLASSMISSING=EXCLUDE /CRITERIA MXITER=100 MXSTEP=5 PCONVERGE=[1E-006 RELATIVE] LCONVERGE=[0] CHKSEP=20 CILEVEL=95 /PRINT SUMMARY VARIABLEINFO SAMPLEINFO HISTORY(1). !doend !enddefine. WEIGHT OFF . CSPLAN ANALYSIS /PLAN FILE='C:\blahblahlah /PLANVARS ANALYSISWEIGHT=svyweight /SRSESTIMATOR TYPE=WR /DESIGN strata= svystrata CLUSTER= svypsu /ESTIMATOR TYPE=WR. !myloop N = 9.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
In reply to this post by E. Bernardo
With regard to comparing two groups with respect to an ordinal categorical variable chi square is insensitive to the ordinal nature of the dependent variable, and effect sizes for the case of nominal variables, such as phi, V, and the contingency coefficient, are not good choices. The Mann-Whitney U test is generally a more powerful test in this case, and the probability-of-superiority ("PS") effect size is generally appropriate. The PS measures the probability that a randomly sampled member of population A will have a higher score (or fall into a higher ordinal category in the present case) than a randomly sampled member of population B. The PS is estimated by U/mn, where m and n are the two sample sizes.
The PS has other names and is discussed in detail in parts of chapters 5 and 9 in: Grissom, Robert J. and Kim. J. J. (2005). Effect sizes for research: A broad practical approach. Mahwah, NJ: Erlbaum. Bob Grissom |
|
|
In reply to this post by E. Bernardo
With regard to comparing two groups with respect to an ordinal categorical variable chi square is insensitive to the ordinal nature of the dependent variable, and effect sizes for the case of nominal variables, such as phi, V, and the contingency coefficient, are not good choices. The Mann-Whitney U test is generally a more powerful test in this case, and the probability-of-superiority ("PS") effect size is generally appropriate. The PS measures the probability that a randomly sampled member of population A will have a higher score (or fall into a higher ordinal category in the present case) than a randomly sampled member of population B. The PS is estimated by U/mn, where m and n are the two sample sizes.
The PS has other names and is discussed in detail in parts of chapters 5 and 9 in: Grissom, Robert J. and Kim. J. J. (2005). Effect sizes for research: A broad practical approach. Mahwah, NJ: Erlbaum. Bob Grissom |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |