Re: st: CR and AVE for factor analysis with 2 factors

[Nick Cox <njcoxstata@gmail.com>]

You just need something like this:

factor zwmdiff_1 zwpdiff_1 zwpdiff_2 zwcost_1 zwcost_2 zwcost_3, pcf
matrix list e(Psi)
matrix list e(L)
matrix psi = e(Psi)'
matrix communalities = J(rowsof(psi),1,1)
matrix communalities = communalities - psi
matrix colnames communalities = communalities
matrix list communalities
mata : st_numscalar("sum_factors", sum(sqrt(st_matrix("communalities"))))
mata : st_numscalar("sum_psi", sum(st_matrix("psi")))
di sum_factors^2/(sum_factors^2+sum_psi)
Nick
njcoxstata@gmail.com


On 29 October 2013 09:49, Florian Christian Esser
<florian.esser@tu-dortmund.de> wrote:
> Hi Nick,
>
> thank you so much! I will try the Mata approach, you suggested, but I have
> to admit that I am not firm in it. In the meantime, would a workaround be
> to set
> gen nvar = e(df_m)
> to
> gen nvar = min(e(df_m),6)
>
> Thank you very much again!
>
>> That explains it then. Your loop goes 1 to 15. You are looking for
>> entries rows 7 to 15 in the matrix, but they don't exist. As soon as
>> you ask for -communalities[7,1]- Stata just overwrites your sum with
>> missing as it evaluates -communalities[7,1]-  as missing. Either your
>> loop must go 1 to 6, in this case, or (if you used the Mata approach I
>> suggested earlier) you need not loop at all.
>> Nick
>> njcoxstata@gmail.com
>>
>>
>> On 29 October 2013 09:33, Florian Christian Esser
>> <florian.esser@tu-dortmund.de> wrote:
>>> It's 15
>>>
>>>> What's e(df_m) in your case?
>>>> Nick
>>>> njcoxstata@gmail.com
>>>>
>>>>
>>>> On 29 October 2013 09:07, Florian Christian Esser
>>>> <florian.esser@tu-dortmund.de> wrote:
>>>>> Here we go (I included the uniqueness and the loadings as well).
>>>>> Communalities are on the bottom:
>>>>> _____________________________________________________________
>>>>>
>>>>>
>>>>> . matrix list e(Psi)
>>>>>
>>>>> e(Psi)[1,6]
>>>>>             zwmdiff_1  zwpdiff_1  zwpdiff_2   zwcost_1   zwcost_2
>>>>> zwcost_3
>>>>> Uniqueness  .14721836  .16061008  .12541414  .01425274  .01366305
>>>>> .02630422
>>>>>
>>>>> . matrix list e(L)
>>>>>
>>>>> e(L)[6,3]
>>>>>               Factor1     Factor2     Factor3
>>>>> zwmdiff_1   .81629684     .233233   .36337787
>>>>> zwpdiff_1   .82275816   .39938175  -.05434294
>>>>> zwpdiff_2    .8671496   .34761446   .04244559
>>>>>  zwcost_1  -.52875169   .83417333   .10160593
>>>>>  zwcost_2  -.51348298   .84223386   .11538764
>>>>>  zwcost_3  -.18959666  -.28021389    .9269461
>>>>>
>>>>> . matrix psi = e(Psi)'
>>>>>
>>>>> . matrix communalities = J(rowsof(psi),1,1)
>>>>>
>>>>> . matrix communalities = communalities - psi
>>>>>
>>>>> . matrix colnames communalities = communalities
>>>>>
>>>>> . matrix list communalities
>>>>>
>>>>> communalities[6,1]
>>>>>            communalit~s
>>>>> zwmdiff_1     .85278164
>>>>> zwpdiff_1     .83938992
>>>>> zwpdiff_2     .87458586
>>>>>  zwcost_1     .98574726
>>>>>  zwcost_2     .98633695
>>>>>  zwcost_3     .97369578
>>>>> _____________________________________________________
>>>>>
>>>>>> What is implied by your log is that
>>>>>>
>>>>>> replace sum_factors = sum_factors + sqrt(communalities[3,1])
>>>>>>
>>>>>> has the effect of replacing -sum_factors- with missing. So, please
>>>>>> show us that -communalities- matrix,
>>>>>> namely the result of your line
>>>>>>
>>>>>> matrix list communalities
>>>>>>
>>>>>> (I see on closer examination that the -communalities- matrix is not
>>>>>> the communalaties as first reported, but modified.)
>>>>>>
>>>>>> No idea, sorry, about your general factor analysis question. I'm
>>>>>> firmly a PCA person.
>>>>>> Nick
>>>>>> njcoxstata@gmail.com
>>>>>>
>>>>>>
>>>>>> On 29 October 2013 07:14, Florian Christian Esser
>>>>>> <florian.esser@tu-dortmund.de> wrote:
>>>>>>> Hi Nick, thanks a lot for your advice.
>>>>>>> Unfortunately, all my communalities are positive.
>>>>>>> Does anyone have any ideas, what else could cause the issue?
>>>>>>>
>>>>>>> Is the general approach correct, to calculate CR and AVE for both
>>>>>>> factors
>>>>>>> at the same time, or would I have to do it individually. I.e. I do
>>>>>>> the
>>>>>>> factor analysis, identify, that the indicators load on two factors
>>>>>>> and
>>>>>>> then calculate CR and AVE for each factor individually?
>>>>>>>
>>>>>>>> Your missing value is presumably the result of taking the square
>>>>>>>> root
>>>>>>>> of a negative number. Is at least one of your communalities
>>>>>>>> reported
>>>>>>>> as negative?
>>>>>>>>
>>>>>>>> That aside you have segments like
>>>>>>>>
>>>>>>>> gen nvar = e(df_m)
>>>>>>>> gen sum_factors=0
>>>>>>>> local i=1
>>>>>>>> while `i' <= nvar {
>>>>>>>> replace sum_factors = sum_factors + sqrt(communalities[`i',1])
>>>>>>>> local i=`i'+1
>>>>>>>> }
>>>>>>>>
>>>>>>>> You can simplify this. First, to hold constants, use locals or
>>>>>>>> scalars, not variables. Second, use -forval- to loop here:
>>>>>>>>
>>>>>>>> local nvar = e(df_m)
>>>>>>>> local sum_factors = 0
>>>>>>>> forval i = 1/`nvar' {
>>>>>>>>      local sum_factors = `sum_factors'  + sqrt(communalities[`i',1]
>>>>>>>> }
>>>>>>>>
>>>>>>>> Third, use Mata instead. Here is a self-contained example:
>>>>>>>>
>>>>>>>> . matrix foo = (1,2,3,4)
>>>>>>>>
>>>>>>>> . mata : st_numscalar("sum_factors", sum(sqrt(st_matrix("foo"))))
>>>>>>>>
>>>>>>>> . scalar li
>>>>>>>> sum_factors =  6.1462644
>>>>>>>>
>>>>>>>> However, none of these tricks can get round what appears to be your
>>>>>>>> problem.
>>>>>>>>
>>>>>>>>
>>>>>>>> Nick
>>>>>>>> njcoxstata@gmail.com
>>>>>>>>
>>>>>>>>
>>>>>>>> On 28 October 2013 14:11, Florian Christian Esser
>>>>>>>> <florian.esser@tu-dortmund.de> wrote:
>>>>>>>>> Hi everyone,
>>>>>>>>>
>>>>>>>>> I am trying to do factor analysis in order to measure two strategy
>>>>>>>>> constructs. I have 6 indicators that load on these two constructs.
>>>>>>>>> Now
>>>>>>>>> I
>>>>>>>>> want to calculate composite reliability (CR) and average variance
>>>>>>>>> extracted(AVE).
>>>>>>>>> I use the following code (here for CR):
>>>>>>>>> ___________________________________________________________________
>>>>>>>>> factor zwmdiff_1 zwpdiff_1 zwpdiff_2 zwcost_1 zwcost_2 zwcost_3,
>>>>>>>>> pcf
>>>>>>>>> matrix list e(Psi)
>>>>>>>>> matrix list e(L)
>>>>>>>>> matrix psi = e(Psi)'
>>>>>>>>> matrix communalities = J(rowsof(psi),1,1)
>>>>>>>>> matrix communalities = communalities - psi
>>>>>>>>> matrix colnames communalities = communalities
>>>>>>>>> matrix list communalities
>>>>>>>>> gen nvar = e(df_m)
>>>>>>>>> gen sum_factors=0
>>>>>>>>> local i=1
>>>>>>>>> while `i' <= nvar {
>>>>>>>>> replace sum_factors = sum_factors + sqrt(communalities[`i',1])
>>>>>>>>> local i=`i'+1
>>>>>>>>> }
>>>>>>>>> generate sum_psi=0
>>>>>>>>> local i=1
>>>>>>>>> while `i' <= nvar {
>>>>>>>>> replace sum_psi = sum_psi + psi[`i',1]
>>>>>>>>> local i=`i'+1
>>>>>>>>> }
>>>>>>>>> gen
>>>>>>>>> cr=(sum_factors*sum_factors)/((sum_factors*sum_factors)+sum_psi)
>>>>>>>>> drop nvar
>>>>>>>>> drop sum_factors
>>>>>>>>> drop sum_psi
>>>>>>>>> list cr in 1
>>>>>>>>> _________________________________________________________________
>>>>>>>>>
>>>>>>>>> The problem is that once the local macro starts running it tells
>>>>>>>>> me:
>>>>>>>>> _____________________________
>>>>>>>>> [...]
>>>>>>>>> (534 real changes made)
>>>>>>>>> (534 real changes made)
>>>>>>>>> (534 real changes made, 534 to missing)
>>>>>>>>> (0 real changes made)
>>>>>>>>> (0 real changes made)
>>>>>>>>> (0 real changes made)
>>>>>>>>> [...]
>>>>>>>>> ___________________________
>>>>>>>>>
>>>>>>>>> and accordingly there is an empty result for CR.
>>>>>>>>> If I add "factors(1)" in the command line above, I get a result,
>>>>>>>>> but
>>>>>>>>> I
>>>>>>>>> since I have two factors, I think it is not correct to do this.
>>>>>>>>>
>>>>>>>>> Does anyone know what to do here?
>>>>>>>>>
>>>>>>>>> Thanks a lot in advance.
>>>>>>>>>
>>>>>>>>> *
>>>>>>>>> *   For searches and help try:
>>>>>>>>> *   http://www.stata.com/help.cgi?search
>>>>>>>>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>>>>>>>>> *   http://www.ats.ucla.edu/stat/stata/
>>>>>>>> *
>>>>>>>> *   For searches and help try:
>>>>>>>> *   http://www.stata.com/help.cgi?search
>>>>>>>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>>>>>>>> *   http://www.ats.ucla.edu/stat/stata/
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> *
>>>>>>> *   For searches and help try:
>>>>>>> *   http://www.stata.com/help.cgi?search
>>>>>>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>>>>>>> *   http://www.ats.ucla.edu/stat/stata/
>>>>>> *
>>>>>> *   For searches and help try:
>>>>>> *   http://www.stata.com/help.cgi?search
>>>>>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>>>>>> *   http://www.ats.ucla.edu/stat/stata/
>>>>>>
>>>>>
>>>>>
>>>>> *
>>>>> *   For searches and help try:
>>>>> *   http://www.stata.com/help.cgi?search
>>>>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>>>>> *   http://www.ats.ucla.edu/stat/stata/
>>>> *
>>>> *   For searches and help try:
>>>> *   http://www.stata.com/help.cgi?search
>>>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>>>> *   http://www.ats.ucla.edu/stat/stata/
>>>>
>>>
>>>
>>> *
>>> *   For searches and help try:
>>> *   http://www.stata.com/help.cgi?search
>>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>>> *   http://www.ats.ucla.edu/stat/stata/
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>> *   http://www.ats.ucla.edu/stat/stata/
>>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/