Positive loadings with PROMAX, negative with OBLIMIN
Locked
Positive loadings with PROMAX, negative with OBLIMIN
 
Administrator
|
Here's another little factor analysis mystery I stumbled across. I'm using the data file available on this UCLA webpage: http://www.ats.ucla.edu/stat/spss/output/factor1.htm. Here's my syntax, with my question at the end.
NEW FILE. DATASET CLOSE all. GET FILE='C:\bw\SPSS\data\UCLA\M255.SAV'. DESCRIPTIVES item13 to item24. factor /variables item13 to item24 /print initial extraction rotation fscore /format blank(.30) /criteria factors(3) iterate(100) /extraction paf /rotation promax /method = correlation. * Factor 3 loadings (from Pattern Matrix) are: 0.781 and 0.821. * This matches results shown in one analysis on the UCLA web-page. factor /variables item13 to item24 /print initial extraction rotation fscore /format blank(.30) /criteria factors(3) iterate(100) /extraction paf /rotation oblimin /method = correlation. * Factor 3 loadings (from Pattern Matrix) are: -0.778 and -0.814. * Q. Why does the sign change when I switch from PROMAX to OBLIMIN? . My apologies to the EFA experts out there if this is a well-known phenomenon. (I'm a bit of a duffer when it comes to multivariate stuff, including EFA!) Cheers, Bruce
--
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/). |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
In any case, there is no intrinsic meaning to the direction that the artificial dimension takes. E.g., for one set of cases to another the signs may all be flipped. Art Kendall Social Research ConsultantsOn 11/8/2013 5:45 PM, Bruce Weaver [via SPSSX Discussion] wrote: Here's another little factor analysis mystery I stumbled across. I'm using the data file available on this UCLA webpage: http://www.ats.ucla.edu/stat/spss/output/factor1.htm. Here's my syntax, with my question at the end.
Art Kendall
Social Research Consultants |
Re: Positive loadings with PROMAX, negative with OBLIMIN
Administrator
|
Bruce,
Plot the loadings in a 3-d scatterplot. The 3rd factor of the oblimin is simply 180 degrees (flipped) from the promax. Sign of the loadings is not of importance WRT interpretation. Do a simple scatter of the two solutions (3rd factor) and they are practically on a straight line (negative slope). I OMS'd the factor and structure matrices and played around but the battery on my LT fizzled before I could wrangle the output. I'll leave it to OP Bruce to drill down into that (best turn off the blank (.30) to explore this). Now, IBM (I guess Jon Why isn't the pattern matrix available in the OMS tags? Oversight? - BTW, why are you the only IBMer that posts here with any regularity? Back in the day a lot of SPSS staff participated. The only person recently was D Dwyer who looked into my MATRIX issue (where are D.Nich.... and D.Math...). Oh well, I guess IBM doesn't give a hoot about SPSSX-L -aside from you Jon) Squeak! HTH, David
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me. --- "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis." Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?" |
|
Just to add a little bit to what David M. says:
(1) The promax oblique rotation starts out with a varimax rotated orthogonal loadings matrix which it then attempts to convert this into a correlated matrix .The SPSS algorithms page goes through the steps here: http://pic.dhe.ibm.com/infocenter/spssstat/v20r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help%2Falg_factor_promax.htm Quoting from the page: The rotated factor pattern is Īpromax=ĪvarimaxQCā1 In the Varimax solution, the loadings for items 23 and 24 are positive and remain positive after the transformation to promax (see the varimax output on the UCLA website). (2) I believe that the oblimin rotation starts out with original unrotated (orthogonal) factor matrix and on the UCLA website, items 23 and 24 have negative loadings on the third factor. The mathematical steps that are involved in doing the oblimin rotation are given on the SPSS algorithms page but some may find the math opaque; see: http://pic.dhe.ibm.com/infocenter/spssstat/v20r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help%2Falg_factor_oblique.htm This point of view is consistent with presentations on the differences between promax and direct oblimin methods; see (esp p435): http://books.google.com/books?hl=en&lr=&id=-VTvN3aPw8sC&oi=fnd&pg=PA424&dq=%22promax+rotation%22+%22factor+analysis%22+%22bipolar+factor%22&ots=Fwr4FO9wUF&sig=f457lVvVf6u76QqYJn0zkHOVD0I#v=onepage&q=oblimin&f=false I believe that in the present case, the oblimin transformation maintains the sign of the original loadings on the third factor though why this is true is not clear to me. While David says that the sign is unimportant I do believe that most analysts would be confused by difference in signs. Also, many report that both promax and oblimin give similar results which one can argue is not the strictly true here. It is possible that someone has studied the conditions under which the signs of the loadings given by promax and oblimin are different but I'm not aware of them. -Mike Palij New York University [hidden email]. ----- Original Message ----- From: "David Marso" <[hidden email]> To: <[hidden email]> Sent: Saturday, November 09, 2013 1:23 AM Subject: Re: Positive loadings with PROMAX, negative with OBLIMIN > Bruce, > Plot the loadings in a 3-d scatterplot. The 3rd factor of the > oblimin is > simply 180 degrees (flipped) from the promax. Sign of the loadings is > not > of importance WRT interpretation. Do a simple scatter of the two > solutions > (3rd factor) and they are practically on a straight line (negative > slope). > I OMS'd the factor and structure matrices and played around but the > battery > on my LT fizzled before I could wrangle the output. > I'll leave it to OP Bruce to drill down into that (best turn off the > blank > (.30) to explore this). Now, IBM (I guess Jon *Why isn't the pattern > matrix > available in the OMS tags? Oversight?* > - BTW, why are you the only IBMer that posts here with any regularity? > Back > in the day a lot of SPSS staff participated. The only person recently > was D > Dwyer who looked into my MATRIX issue (where are D.Nich.... and > D.Math...). > Oh well, I guess IBM doesn't give a hoot about SPSSX-L -aside from you > Jon) > Squeak! > HTH, David > > Art Kendall wrote >> In any case, there is no intrinsic meaning to the direction that >> the artificial dimension takes. E.g., for one set of >> cases >> to >> another the signs may all be flipped. >> >> >> Art Kendall >> Social Research Consultants >> On 11/8/2013 5:45 PM, Bruce Weaver [via SPSSX Discussion] >> wrote: >> >> Here's another little factor analysis mystery I >> stumbled across. I'm using the data file available on >> this >> UCLA >> webpage: >> http://www.ats.ucla.edu/stat/spss/output/factor1.htm >> . >> Here's my syntax, with my question at the end. >> >> >> >> NEW FILE. >> >> DATASET CLOSE all. >> >> >> GET FILE='C:\bw\SPSS\data\UCLA\M255.SAV'. >> >> >> DESCRIPTIVES item13 to item24. >> >> >> factor >> >> /variables item13 to item24 >> >> /print initial extraction rotation fscore >> >> /format blank(.30) >> >> /criteria factors(3) iterate(100) >> >> /extraction paf >> >> /rotation promax >> >> /method = correlation. >> >> >> * Factor 3 loadings (from Pattern Matrix) are: 0.781 and >> 0.821. >> >> * This matches results shown in one analysis on the UCLA >> web-page. >> >> >> factor >> >> /variables item13 to item24 >> >> /print initial extraction rotation fscore >> >> /format blank(.30) >> >> /criteria factors(3) iterate(100) >> >> /extraction paf >> >> /rotation oblimin >> >> /method = correlation. >> >> >> * Factor 3 loadings (from Pattern Matrix) are: -0.778 and >> -0.814. >> >> >> * Q. Why does the sign change when I switch from PROMAX to >> OBLIMIN? . >> >> >> >> My apologies to the EFA experts out there if this is a >> well-known >> phenomenon. (I'm a bit of a duffer when it comes to >> multivariate >> stuff, including EFA!) >> >> >> Cheers, >> >> Bruce >> >> >> -- >> >> Bruce Weaver >> >> > >> bweaver@ > >> >> >> http://sites.google.com/a/lakeheadu.ca/bweaver/ >> >> "When all else fails, RTFM." >> >> >> NOTE: My Hotmail account is not monitored regularly. >> >> To send me an e-mail, please use the address shown above. >> >> >> >> >> >> >> If you reply to this email, your >> message will be added to the discussion below: >> >> http://spssx-discussion.1045642.n5.nabble.com/Positive-loadings-with-PROMAX-negative-with-OBLIMIN-tp5722956.html >> >> >> To start a new topic under SPSSX Discussion, email >> > >> ml-node+s1045642n1068821h68@.nabble > >> >> To unsubscribe from SPSSX Discussion, click >> here . >> NAML > > > > > > ----- > Please reply to the list and not to my personal email. > Those desiring my consulting or training services please feel free to > email me. > --- > "Nolite dare sanctum canibus neque mittatis margaritas vestras ante > porcos ne forte conculcent eas pedibus suis." > Cum es damnatorum possederunt porcos iens ut salire off sanguinum > cliff in abyssum?" > -- > View this message in context: > http://spssx-discussion.1045642.n5.nabble.com/Positive-loadings-with-PROMAX-negative-with-OBLIMIN-tp5722956p5722958.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 |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
Administrator
|
Thanks to Art, David & Mike. That's helpful.
--
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/). |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
|
In reply to this post by David Marso
Now, IBM (I guess Jon *Why isn't the pattern matrix
available in the OMS tags? Oversight?* >>>Every table from FACTOR is available in OMS. Which table type did you produce that you didn't see listed - right click in the outline on the title and choose Copy OMS table subtype? - BTW, why are you the only IBMer that posts here with any regularity? >>>SPSS TS folks do monitor the list, but they also respond on the SPSS Community forums. Senior Software Engineer, IBM [hidden email] phone: 720-342-5621 |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
Administrator
|
My bad: I was looking for Pattern Matrix: The actual OMS subtype is 'Rotated Factor Matrix'
I didn't know about right click for the subtype! Thanks Jon. ---
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me. --- "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis." Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?" |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
|
In reply to this post by Bruce Weaver
[re-post, in case you did see the first. Slightly re-worded.]
As everyone says, the +/- direction of the vector is wholly arbitrary and irrelevant to the meaning. A method of solution results in a set of signs; a slight variation in the method (or data) might readily result in reversed signs. I remember a convention suggested for how-to-display solutions: compare the sums of negative and positive loading-squared terms, and switch all signs if the negatives are bigger. That could be a standard either for programming or for data presentation in reports. "Eyeballing" usually gives an easy answer for Varimax. Oblique rotations give larger loadings in general. Simplified standards for eyeballing include, "Make the largest loading positive" and "Make the majority of loadings positive." I wonder, but not very much, how many computer programs implement any of these rules of thumb. -- Rich Ulrich > Date: Sat, 9 Nov 2013 05:46:03 -0800 > From: [hidden email] > Subject: Re: Positive loadings with PROMAX, negative with OBLIMIN > To: [hidden email] > > Thanks to Art, David & Mike. That's helpful. > > ... |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
|
The signs on the factor
loading are completely arbitrary.
Rich's suggestions are certainly good ways to go in many circumstances. When one wishes to interpret the factor solution and especially when one wants to develop a scoring key, e.g., for values, attitudes, or solutions that end up with Osgood' semantic differential dimensions, I would use semantic criteria to name and pick the direction according to what the underlying construct is interpreted to be. I would reflect items so the resulting score would go from semantically negative to positive (left to right or down to up when graphing) low ... high weak ... strong no threat ... very threatening undesirable ... desirable quiet ... active inexpensive ... expensive rarely ... frequently unfair ... fair inefficient ... efficient Art Kendall Social Research ConsultantsOn 11/9/2013 11:33 AM, Rich Ulrich [via SPSSX Discussion] wrote:
Art Kendall
Social Research Consultants |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
|
Art,
Good points, entirely. I suppose my recommendation suits for the step, "What does the data analyst first present to the PI?" The PI needs to be made aware that he can use either end of the scale. What makes the best name to represent this dimension? Reverse the signs, if that gives you the better name for representing the scores. Rich Ulrich Date: Sat, 9 Nov 2013 10:21:40 -0800 From: [hidden email] Subject: Re: Positive loadings with PROMAX, negative with OBLIMIN To: [hidden email] The signs on the factor
loading are completely arbitrary.
Rich's suggestions are certainly good ways to go in many circumstances. When one wishes to interpret the factor solution and especially when one wants to develop a scoring key, e.g., for values, attitudes, or solutions that end up with Osgood' semantic differential dimensions, I would use semantic criteria to name and pick the direction according to what the underlying construct is interpreted to be. I would reflect items so the resulting score would go from semantically negative to positive (left to right or down to up when graphing) low ... high weak ... strong no threat ... very threatening undesirable ... desirable quiet ... active inexpensive ... expensive rarely ... frequently unfair ... fair inefficient ... efficient Art Kendall Social Research ConsultantsOn 11/9/2013 11:33 AM, Rich Ulrich [via SPSSX Discussion] wrote:
|
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
|
Yes your recommendation
does suit for the first look at the loadings. In fact, I
habitually try to think or both orders of items on a factor in
the process of finding a name to attach to the underlying
construct.
Art Kendall Social Research ConsultantsOn 11/10/2013 12:41 AM, Rich Ulrich [via SPSSX Discussion] wrote:
Art Kendall
Social Research Consultants |
Re: Positive loadings with PROMAX, negative with OBLIMIN
|
|
In reply to this post by Bruce Weaver
Hello everyone.
I have a similar `problemĀ“ (I work with SPSS 22, not sure if this is relevant). I get 14 factors with high positive loadings with varimax but with an oblimin rotation I get 14 factors where the items load in the same pattern, but they have negative or positive loadings (each factor has Items with either positive or negative loadings). The answers, that the sign is irrelevant helps me, but is there any literature to confirm this for a master thesis? Thanks a lot |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |