[R] Help with factanal and missing values
Hi list, I'm performing a series of confirmatory factor analysis on different groupings of items from data collected with questionnaires. There are some missing values. For those sets with no missing values I call factanal(datamatrix,factors=n) where datamatrix is a table of all observations for the items under investigation. This call fails when there are missing values. help(factanal) does not give an example on calls with na.action and and mentiones a formula. (Venables and Ripley, 2002 give only one example on p. 323 for a case where the covariance has already been calculated) Could someone give me an example on such a call for a simple CFA? Thanks for the help. Saluti, Antonio Prioglio -- We are what we repeatedly do. Excellence, then, is not an act, but a habit. Aristoteles /\ \ /ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL X - AGAINST MS ATTACHMENTS / \ http://www.gnu.org/philosophy/no-word-attachments.html __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
Dear Antonio, -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Antonio Prioglio Sent: Tuesday, July 13, 2004 9:32 AM To: [EMAIL PROTECTED] Subject: [R] Help with factanal and missing values Hi list, I'm performing a series of confirmatory factor analysis on different groupings of items from data collected with questionnaires. There are some missing values. For those sets with no missing values I call factanal(datamatrix,factors=n) where datamatrix is a table of all observations for the items under investigation. This call fails when there are missing values. help(factanal) does not give an example on calls with na.action and and mentiones a formula. (Venables and Ripley, 2002 give only one example on p. 323 for a case where the covariance has already been calculated) Could someone give me an example on such a call for a simple CFA? Two solutions are to use na.omit() to eliminate observations with missing data -- factanal(na.omit(datamatrix), factors=n) -- or to use a formula argument to factanal and pass the data as a data frame via the data argument -- factanal(~ var1 + ... + vark, factors=n, data=as.data.frame(datamatrix)). By the way, what factanal() does would conventionally be described as exploratory, not confirmatory, factor analysis. For the latter, you might try the sem package. I hope that this helps, John __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
On Tue, 13 Jul 2004, John Fox wrote: Two solutions are to use na.omit() to eliminate observations with missing data -- factanal(na.omit(datamatrix), factors=n) -- or to use a formula argument to factanal and pass the data as a data frame via the data argument -- factanal(~ var1 + ... + vark, factors=n, data=as.data.frame(datamatrix)). Thanks this was helpful and more elegant than my solution to select the data with a call to complete.cases(). By the way, what factanal() does would conventionally be described as exploratory, not confirmatory, factor analysis. For the latter, you might try the sem package. Actually these CFAs are preliminary to a Path Analysis I had ambitions to do with the sem package rather than LISREL. Could you give an example on how to do CFA with sem? Saluti, Antonio Prioglio -- We are what we repeatedly do. Excellence, then, is not an act, but a habit. Aristoteles /\ \ /ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL X - AGAINST MS ATTACHMENTS / \ http://www.gnu.org/philosophy/no-word-attachments.html __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
Dear Antonio, -Original Message- From: Antonio Prioglio [mailto:[EMAIL PROTECTED] Sent: Tuesday, July 13, 2004 12:50 PM To: John Fox Cc: [EMAIL PROTECTED] Subject: RE: [R] Help with factanal and missing values On Tue, 13 Jul 2004, John Fox wrote: Two solutions are to use na.omit() to eliminate observations with missing data -- factanal(na.omit(datamatrix), factors=n) -- or to use a formula argument to factanal and pass the data as a data frame via the data argument -- factanal(~ var1 + ... + vark, factors=n, data=as.data.frame(datamatrix)). Thanks this was helpful and more elegant than my solution to select the data with a call to complete.cases(). By the way, what factanal() does would conventionally be described as exploratory, not confirmatory, factor analysis. For the latter, you might try the sem package. Actually these CFAs are preliminary to a Path Analysis I had ambitions to do with the sem package rather than LISREL. Could you give an example on how to do CFA with sem? Among the examples in ?sem is a second-order CFA. Regards, John __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
On Tue, 13 Jul 2004, John Fox wrote: Could you give an example on how to do CFA with sem? Among the examples in ?sem is a second-order CFA. Ok, sorry I forgot about the Thurstone example as I was focusing on Path Analysis example when reading the sem doc. I understand this is a model taken from another source. In the model specification I notice a F4-F1, F4-F2, F4-F3. This is something I do not understand. In a normal case where one is iterested in the factorial validity of the indicator variables, would it be appropiate to use the same specification just omitting the above links? Saluti, Antonio Prioglio -- We are what we repeatedly do. Excellence, then, is not an act, but a habit. Aristoteles /\ \ /ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL X - AGAINST MS ATTACHMENTS / \ http://www.gnu.org/philosophy/no-word-attachments.html __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
On Tue, 13 Jul 2004, John Fox wrote: Among the examples in ?sem is a second-order CFA. Thanks now I have it running, hopefully in a correct way. By the way I notice in the output a GFI index that in your online appendix to your 2002 book describe as an ad hoc measure. Could you comment on how it compares with the Comparative Fit Index (CFI; Bentler, 1990) which according to Byrne 1994 is a revised version of the Bentler-Bonnet (1980) Normed Fit Index? Thanks again for your valuable help. Saluti, Antonio Prioglio -- We are what we repeatedly do. Excellence, then, is not an act, but a habit. Aristoteles /\ \ /ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL X - AGAINST MS ATTACHMENTS / \ http://www.gnu.org/philosophy/no-word-attachments.html __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
Dear Antonio, This example is (as stated) a second-order CFI, where each of the primary factors, F1, F2, and F3 depends upon the second-order factor F4. To have no second-order structure, simply define variances and (assuming that you're specifying correlated factors) covariances among the factors. For the Thurstone example, these would be, in addition to the variances already set to 1, F1 - F2, F1 - F3, and F2 - F3. I hope that this helps, John -Original Message- From: Antonio Prioglio [mailto:[EMAIL PROTECTED] Sent: Tuesday, July 13, 2004 1:53 PM To: John Fox Cc: [EMAIL PROTECTED] Subject: RE: [R] Help with factanal and missing values On Tue, 13 Jul 2004, John Fox wrote: Could you give an example on how to do CFA with sem? Among the examples in ?sem is a second-order CFA. Ok, sorry I forgot about the Thurstone example as I was focusing on Path Analysis example when reading the sem doc. I understand this is a model taken from another source. In the model specification I notice a F4-F1, F4-F2, F4-F3. This is something I do not understand. In a normal case where one is iterested in the factorial validity of the indicator variables, would it be appropiate to use the same specification just omitting the above links? Saluti, Antonio Prioglio -- We are what we repeatedly do. Excellence, then, is not an act, but a habit. Aristoteles /\ \ /ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL X - AGAINST MS ATTACHMENTS / \ http://www.gnu.org/philosophy/no-word-attachments.html __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
Dear Antonio, I'm afraid that you're asking the wrong person about this, because the literature on goodness of fit measures in SEMs strikes me as mostly alchemy. As I recall, there are several chapters discussing fit measures in Bollen and Long, eds., Testing Structural Equation Models. Perhaps these would be of help. Regards, John -Original Message- From: Antonio Prioglio [mailto:[EMAIL PROTECTED] Sent: Tuesday, July 13, 2004 3:22 PM To: John Fox Cc: [EMAIL PROTECTED] Subject: RE: [R] Help with factanal and missing values On Tue, 13 Jul 2004, John Fox wrote: Among the examples in ?sem is a second-order CFA. Thanks now I have it running, hopefully in a correct way. By the way I notice in the output a GFI index that in your online appendix to your 2002 book describe as an ad hoc measure. Could you comment on how it compares with the Comparative Fit Index (CFI; Bentler, 1990) which according to Byrne 1994 is a revised version of the Bentler-Bonnet (1980) Normed Fit Index? Thanks again for your valuable help. __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
On Tue, 13 Jul 2004, John Fox wrote: Dear Antonio, This example is (as stated) a second-order CFI, where each of the primary factors, F1, F2, and F3 depends upon the second-order factor F4. To have no second-order structure, simply define variances and (assuming that you're specifying correlated factors) covariances among the factors. For the Thurstone example, these would be, in addition to the variances already set to 1, F1 - F2, F1 - F3, and F2 - F3. Ok I understand the second order part in relation to F4. I'm afraid I'm a bit slow on the last part. Fn - Fn NA 1 is the variance for the factor n. Fn - Item.a lam.z NA is the grouping of the items to each factor Fn - Fm gam.w NA I take is the covariance among factors. Is this correct? Do I need to set variances for each item (Item.a - Item.a)? Saluti, Antonio Prioglio -- We are what we repeatedly do. Excellence, then, is not an act, but a habit. Aristoteles /\ \ /ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL X - AGAINST MS ATTACHMENTS / \ http://www.gnu.org/philosophy/no-word-attachments.html __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
RE: [R] Help with factanal and missing values
Dear Antonio, -Original Message- From: Antonio Prioglio [mailto:[EMAIL PROTECTED] Sent: Tuesday, July 13, 2004 5:54 PM To: John Fox Cc: [EMAIL PROTECTED] Subject: RE: [R] Help with factanal and missing values On Tue, 13 Jul 2004, John Fox wrote: Dear Antonio, This example is (as stated) a second-order CFI, where each of the primary factors, F1, F2, and F3 depends upon the second-order factor F4. To have no second-order structure, simply define variances and (assuming that you're specifying correlated factors) covariances among the factors. For the Thurstone example, these would be, in addition to the variances already set to 1, F1 - F2, F1 - F3, and F2 - F3. Ok I understand the second order part in relation to F4. I'm afraid I'm a bit slow on the last part. Fn - FnNA 1 is the variance for the factor n. Yes. This is set to 1, standardizing the factors, which is conventional. Fn - Item.a lam.z NA is the grouping of the items to each factor These (i.e., the lambdas) are conventionally called factor loadings. Fn - Fm gam.w NA I take is the covariance among factors. Yes, though it is unconventional to use gamma to represent a covariance (a correlation, since the factors are presumably standardized) in factor analysis. Is this correct? Do I need to set variances for each item (Item.a - Item.a)? Yes. These *error* variances, which should be free parameters, are conventionally called uniquenesses in factor analysis. As a general matter, it would help, I think, to look more carefully at the example, since it includes factor loadings and uniquenesses. If you haven't already done so, you might also look at the appendix on structural-equation models to my R and S-PLUS Companion, which is at http://socserv.socsci.mcmaster.ca/jfox/Books/Companion/appendix-sems.pdf. This appendix doesn't include a CFA example, but it does explain how to prepare input to the sem() function. Regards, John Saluti, Antonio Prioglio -- We are what we repeatedly do. Excellence, then, is not an act, but a habit. Aristoteles /\ \ /ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL X - AGAINST MS ATTACHMENTS / \ http://www.gnu.org/philosophy/no-word-attachments.html __ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
