Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Syntax for a mixed-effect repeated measures model

    Dear Statalisters,

    I am trying to implement a mixed-effet repeated measures model that was described by Hamlett, Ryan, Serrano-Trespalacios & Wolfinger in their paper entitled "" (Journal of Air and Waste Management Association, 2003. 53:4 442-50). The paper appears to be freely accessible from the publisher.

    The model is described mathematically, where repeated measures are taken on individuals using two different, correlated, measurements. They then give a worked example of implementing the model in SAS. From reading the article, I know there should be a random-effect for the measurement type with an unstructured covariance matrix, and then residual errors of each replicate to be nested within each individual. I can't seem to wrap my head around the appropriate syntax for -mixed-.

    I've provided a data example below. My best guess at the syntax would be:

    Code:
    mixed response ibn.vtype, nocons || person : vtype, cov(un) || _all : , nocons resid(ind, by(vtype) t(repl))
    These data are for the linked data example and are attributable to Bland and Altman.

    Code:
    * Example generated by -dataex-. To install: ssc install dataex
    clear
    input byte(personid vtype replicate) double(response)
    1 1 1 6.68
    1 2 1 3.97
    1 1 2 6.53
    1 2 2 4.12
    1 1 3 6.43
    1 2 3 4.09
    1 1 4 6.33
    1 2 4 3.97
    2 1 1 6.85
    2 2 1 5.27
    2 1 2 7.06
    2 2 2 5.37
    2 1 3 7.13
    2 2 3 5.41
    2 1 4 7.17
    2 2 4 5.44
    3 1 1  7.4
    3 2 1 5.67
    3 1 2 7.42
    3 2 2 3.64
    3 1 3 7.41
    3 2 3 4.32
    3 1 4 7.37
    3 2 4 4.73
    3 1 5 7.34
    3 2 5 4.96
    3 1 6 7.35
    3 2 6 5.04
    3 1 7 7.28
    3 2 7 5.22
    3 1 8  7.3
    3 2 8 4.82
    3 1 9 7.34
    3 2 9 5.07
    4 1 1 7.36
    4 2 1 5.67
    4 1 2 7.33
    4 2 2  5.1
    4 1 3 7.29
    4 2 3 5.53
    4 1 4  7.3
    4 2 4 4.75
    4 1 5 7.35
    4 2 5 5.51
    5 1 1 7.35
    5 2 1 4.28
    5 1 2  7.3
    5 2 2 4.44
    5 1 3  7.3
    5 2 3 4.32
    5 1 4 7.37
    5 2 4 3.23
    5 1 5 7.27
    5 2 5 4.46
    5 1 6 7.28
    5 2 6 4.72
    5 1 7 7.32
    5 2 7 4.75
    5 1 8 7.32
    5 2 8 4.99
    6 1 1 7.38
    6 2 1 4.78
    6 1 2  7.3
    6 2 2 4.73
    6 1 3 7.29
    6 2 3 5.12
    6 1 4 7.33
    6 2 4 4.93
    6 1 5 7.31
    6 2 5 5.03
    6 1 6 7.33
    6 2 6 4.93
    7 1 1 6.86
    7 2 1 6.85
    7 1 2 6.94
    7 2 2 6.44
    7 1 3 6.92
    7 2 3 6.52
    8 1 1 7.19
    8 2 1 5.28
    8 1 2 7.29
    8 2 2 4.56
    8 1 3 7.21
    8 2 3 4.34
    8 1 4 7.25
    8 2 4 4.32
    8 1 5  7.2
    8 2 5 4.41
    8 1 6 7.19
    8 2 6 3.69
    8 1 7 6.77
    8 2 7 6.09
    8 1 8 6.82
    8 2 8 5.58
    end
    Any help is much appreciated.

  • #2
    Here's my take on it, which reproduces the results in the paper's two main tables for the example (Tables 5 and 8).

    .ÿ
    .ÿversionÿ15.1

    .ÿ
    .ÿclearÿ*

    .ÿ
    .ÿquietlyÿinputÿbyte(personidÿvtypeÿreplicate)ÿdouble(response)

    .ÿ
    .ÿquietlyÿtabulateÿvtype,ÿgenerate(v)

    .ÿ
    .ÿ//ÿPaper'sÿTableÿ5ÿ(alsoÿµxÿandÿµyÿinÿTableÿ4)
    .ÿmixedÿresponseÿibn.vtype,ÿnoconstantÿ///
    >ÿÿÿÿÿÿÿÿÿ||ÿpersonid:ÿv1ÿv2,ÿnoconstantÿcovariance(unstructured)ÿ///
    >ÿÿÿÿÿÿÿÿÿ||ÿreplicate:ÿ,ÿnoconstantÿresiduals(unstructured,ÿt(vtype))ÿ///
    >ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿmleÿnogroupÿnolrtestÿnolog

    Mixed-effectsÿMLÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿÿÿÿÿ94

    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿWaldÿchi2(2)ÿÿÿÿÿÿ=ÿÿÿÿ5438.14
    Logÿlikelihoodÿ=ÿ-14.224648ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿchi2ÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0000

    ------------------------------------------------------------------------------
    ÿÿÿÿresponseÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
    -------------+----------------------------------------------------------------
    ÿÿÿÿÿÿÿvtypeÿ|
    ÿÿÿÿÿÿÿÿÿÿ1ÿÿ|ÿÿÿ7.115019ÿÿÿ.0980337ÿÿÿÿ72.58ÿÿÿ0.000ÿÿÿÿÿ6.922877ÿÿÿÿ7.307162
    ÿÿÿÿÿÿÿÿÿÿ2ÿÿ|ÿÿÿ5.008169ÿÿÿ.2434462ÿÿÿÿ20.57ÿÿÿ0.000ÿÿÿÿÿ4.531023ÿÿÿÿ5.485315
    ------------------------------------------------------------------------------

    ------------------------------------------------------------------------------
    ÿÿRandom-effectsÿParametersÿÿ|ÿÿÿEstimateÿÿÿStd.ÿErr.ÿÿÿÿÿ[95%ÿConf.ÿInterval]
    -----------------------------+------------------------------------------------
    personid:ÿUnstructuredÿÿÿÿÿÿÿ|
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(v1)ÿ|ÿÿÿ.0746488ÿÿÿ.0386207ÿÿÿÿÿÿ.0270794ÿÿÿÿ.2057811
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(v2)ÿ|ÿÿÿ.4252481ÿÿÿ.2493731ÿÿÿÿÿÿ.1347357ÿÿÿÿ1.342154
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿcov(v1,v2)ÿ|ÿÿÿ.0252172ÿÿÿÿ.068893ÿÿÿÿÿ-.1098106ÿÿÿÿÿ.160245
    -----------------------------+------------------------------------------------
    replicate:ÿÿÿÿÿÿÿÿÿÿÿ(empty)ÿ|
    -----------------------------+------------------------------------------------
    Residual:ÿUnstructuredÿÿÿÿÿÿÿ|
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(e1)ÿ|ÿÿÿ.0115598ÿÿÿ.0026283ÿÿÿÿÿÿÿ.007403ÿÿÿÿ.0180504
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(e2)ÿ|ÿÿÿ.2546715ÿÿÿÿ.058312ÿÿÿÿÿÿ.1625852ÿÿÿÿ.3989143
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿcov(e1,e2)ÿ|ÿÿ-.0276256ÿÿÿ.0098326ÿÿÿÿÿ-.0468971ÿÿÿ-.0083541
    ------------------------------------------------------------------------------

    .ÿ
    .ÿ//ÿMeansÿandÿt-statisticsÿinÿpaper'sÿTableÿ8
    .ÿmixedÿresponseÿib(last).vtypeÿ///
    >ÿÿÿÿÿÿÿÿÿ||ÿpersonid:ÿv1ÿv2,ÿnoconstantÿcovariance(unstructured)ÿ///
    >ÿÿÿÿÿÿÿÿÿ||ÿreplicate:ÿ,ÿnoconstantÿresiduals(unstructured,ÿt(vtype))ÿ///ÿremlÿdfmethod(kroger)ÿ
    >ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿdfmethod(anova)ÿnoretableÿnogroupÿnolrtestÿnolog

    Mixed-effectsÿMLÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿÿÿÿÿ94
    DFÿmethod:ÿANOVAÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿDF:ÿÿÿÿÿÿÿÿÿÿÿminÿ=ÿÿÿÿÿÿ78.00
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿavgÿ=ÿÿÿÿÿÿ78.00
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿmaxÿ=ÿÿÿÿÿÿ78.00

    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿF(1,ÿÿÿÿ78.00)ÿÿÿÿ=ÿÿÿÿÿÿ69.47
    Logÿlikelihoodÿ=ÿ-14.224648ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿÿÿÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0000

    ------------------------------------------------------------------------------
    ÿÿÿÿresponseÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿtÿÿÿÿP>|t|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
    -------------+----------------------------------------------------------------
    ÿÿÿÿÿ1.vtypeÿ|ÿÿÿ2.106851ÿÿÿ.2527693ÿÿÿÿÿ8.34ÿÿÿ0.000ÿÿÿÿÿ1.603626ÿÿÿÿ2.610076
    ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ5.008169ÿÿÿ.2434462ÿÿÿÿ20.57ÿÿÿ0.000ÿÿÿÿÿ4.523505ÿÿÿÿ5.492833
    ------------------------------------------------------------------------------

    .ÿ
    .ÿ
    .ÿexit

    endÿofÿdo-file


    .


    I haven't bothered wth Tables 6 and 7; I assume that you can get those, too, if you want, with some estat postestimation command or other.

    It seems that SAS's implementation of the Kenward-Roger adjustment at least then differs from how Stata implements it, because the regression coefficients and their standard errors, as well as the denominator degrees of freedom shown in Table 8 there differ from what you'd get with that dfmethod() in the set up I've shown above.

    Comment


    • #3
      Dear , thank you very much for taking the time to help me replicate this analysis. I think the key stumbling blocks I had were not explicitly nesting replications in persons, and also the neat trick of parameterizing each vtype as their own indicator variables, v1 and v2.

      For anyone who is interested in reading this, the remaining results from that paper can be derived using -estat- post-estimation as suggested. Secondly, while they still do differ after about the 3rd or 4th decimal point, Stata uses the expected information matrix whereas SAS uses the observed information matrix, by default, for computing the Kenward-Rodger df method.

      Code:
      . // estimated covariance matrix
      . estat wcorr, cov at(person=1) format(%9.6f)
      
      Covariances for personid = 1:
      
             vtype |         1          2          1          2          1          2          1          2
      -------------+----------------------------------------------------------------------------------------
                 1 |  0.086209                                                                              
                 2 | -0.002408   0.679920                                                                   
                 1 |  0.074649   0.025217   0.086209                                                        
                 2 |  0.025217   0.425248  -0.002408   0.679920                                             
                 1 |  0.074649   0.025217   0.074649   0.025217   0.086209                                  
                 2 |  0.025217   0.425248   0.025217   0.425248  -0.002408   0.679920                       
                 1 |  0.074649   0.025217   0.074649   0.025217   0.074649   0.025217   0.086209            
                 2 |  0.025217   0.425248   0.025217   0.425248   0.025217   0.425248  -0.002408   0.679920
      
      . // estimated correlation matrix
      . estat wcorr, at(person=1) format(%9.6f)
      
      Standard deviations and correlations for personid = 1:
      
      Correlations:
      
             vtype |         1          2          1          2          1          2          1          2
      -------------+----------------------------------------------------------------------------------------
                 1 |  1.000000                                                                              
                 2 | -0.009948   1.000000                                                                   
                 1 |  0.865909   0.104158   1.000000                                                        
                 2 |  0.104158   0.625439  -0.009948   1.000000                                             
                 1 |  0.865909   0.104158   0.865909   0.104158   1.000000                                  
                 2 |  0.104158   0.625439   0.104158   0.625439  -0.009948   1.000000                       
                 1 |  0.865909   0.104158   0.865909   0.104158   0.865909   0.104158   1.000000            
                 2 |  0.104158   0.625439   0.104158   0.625439   0.104158   0.625439  -0.009948   1.000000
      
      . // estimated covariance matrix as the sum of R and G matrices
      . estat wcorr, cov format(%9.6f)
      
      Covariances for personid = 1 replicate = 1:
      
             vtype |         1          2
      -------------+----------------------
                 1 |  0.086209            
                 2 | -0.002408   0.679920

      Comment


      • #4
        Originally posted by Leonardo Guizzetti View Post
        Secondly, while they still do differ after about the 3rd or 4th decimal point, Stata uses the expected information matrix whereas SAS uses the observed information matrix, by default, for computing the Kenward-Rodger df method.
        I had already checked that and ruled it out as the cause of the discrepancy before posting. Results are quite a bit different from what is shown in the paper's Table 8. See below. You got something different using SAS yourself from what they report there?

        Anyway, because the research focus is on correlations between paired measurements, if it were me, I would have initially approached this from an SEM perspective, which seems to me to be a more natural fit to the objective. I know that the answers are the same, but I would have expected not to have to think so much about how to set up the model (variances and covariances) in gsem as with mixed (or PROC MIXED).

        .ÿ
        .ÿversionÿ15.1

        .ÿ
        .ÿclearÿ*

        .ÿ
        .ÿquietlyÿinputÿbyte(personidÿvtypeÿreplicate)ÿdouble(response)

        .ÿ
        .ÿquietlyÿtabulateÿvtype,ÿgenerate(v)

        .ÿ
        .ÿmixedÿresponseÿib(last).vtypeÿ///
        >ÿÿÿÿÿÿÿÿÿ||ÿpersonid:ÿv1ÿv2,ÿnoconstantÿcovariance(unstructured)ÿ///
        >ÿÿÿÿÿÿÿÿÿ||ÿreplicate:ÿ,ÿnoconstantÿresiduals(unstructured,ÿt(vtype))ÿ///
        >ÿÿÿÿÿÿÿÿÿremlÿdfmethod(kroger,ÿoim)ÿnogroupÿnolrtestÿnoretableÿnolog

        Mixed-effectsÿREMLÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿÿÿÿÿ94
        DFÿmethod:ÿKenward-RogerÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿDF:ÿÿÿÿÿÿÿÿÿÿÿminÿ=ÿÿÿÿÿÿÿ6.36
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿavgÿ=ÿÿÿÿÿÿÿ6.39
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿmaxÿ=ÿÿÿÿÿÿÿ6.42

        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿF(1,ÿÿÿÿÿ6.36)ÿÿÿÿ=ÿÿÿÿÿÿ59.51
        Logÿrestricted-likelihoodÿ=ÿ-16.058561ÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿÿÿÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0002

        ------------------------------------------------------------------------------
        ÿÿÿÿresponseÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿtÿÿÿÿP>|t|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
        -------------+----------------------------------------------------------------
        ÿÿÿÿÿ1.vtypeÿ|ÿÿÿÿ2.10307ÿÿÿ.2726266ÿÿÿÿÿ7.71ÿÿÿ0.000ÿÿÿÿÿ1.444996ÿÿÿÿ2.761145
        ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ5.011394ÿÿÿÿ.262403ÿÿÿÿ19.10ÿÿÿ0.000ÿÿÿÿÿ4.379336ÿÿÿÿ5.643453
        ------------------------------------------------------------------------------

        .ÿ
        .ÿdropÿv?

        .ÿquietlyÿreshapeÿwideÿresponse,ÿi(personidÿreplicate)ÿj(vtype)

        .ÿgsemÿ///
        >ÿÿÿÿÿÿÿÿÿ(response1ÿ<-ÿM1[personid])ÿ///
        >ÿÿÿÿÿÿÿÿÿ(response2ÿ<-ÿM2[personid]),ÿ///
        >ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿcovariance(M1[personid]*M2[personid]ÿe.response1*e.response2)ÿ///
        >ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿnocnsreportÿnodvheaderÿnolog

        GeneralizedÿstructuralÿequationÿmodelÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿÿÿÿÿ47
        Logÿlikelihoodÿ=ÿ-14.224648

        ------------------------------------------------------------------------------------------------
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
        -------------------------------+----------------------------------------------------------------
        response1ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿM1[personid]ÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿ(constrained)
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ_consÿ|ÿÿÿ7.115019ÿÿÿ.0980632ÿÿÿÿ72.56ÿÿÿ0.000ÿÿÿÿÿ6.922819ÿÿÿÿÿ7.30722
        -------------------------------+----------------------------------------------------------------
        response2ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿM2[personid]ÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿ(constrained)
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ_consÿ|ÿÿÿ5.008169ÿÿÿ.2438641ÿÿÿÿ20.54ÿÿÿ0.000ÿÿÿÿÿ4.530204ÿÿÿÿ5.486134
        -------------------------------+----------------------------------------------------------------
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(M1[personid])|ÿÿÿ.0746488ÿÿÿ.0386207ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.0270794ÿÿÿÿ.2057811
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(M2[personid])|ÿÿÿ.4252481ÿÿÿ.2493731ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.1347357ÿÿÿÿ1.342153
        -------------------------------+----------------------------------------------------------------
        ÿcov(M1[personid],M2[personid])|ÿÿÿ.0252172ÿÿÿÿ.068893ÿÿÿÿÿ0.37ÿÿÿ0.714ÿÿÿÿ-.1098107ÿÿÿÿÿ.160245
        -------------------------------+----------------------------------------------------------------
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(e.response1)|ÿÿÿ.0115598ÿÿÿ.0026283ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.0074031ÿÿÿÿ.0180504
        ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(e.response2)|ÿÿÿ.2546715ÿÿÿÿ.058312ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.1625852ÿÿÿÿ.3989143
        -------------------------------+----------------------------------------------------------------
        ÿÿÿcov(e.response1,e.response2)|ÿÿ-.0276256ÿÿÿ.0098326ÿÿÿÿ-2.81ÿÿÿ0.005ÿÿÿÿ-.0468971ÿÿÿ-.0083541
        ------------------------------------------------------------------------------------------------

        .ÿ
        .ÿexit

        endÿofÿdo-file


        .

        Comment


        • #5
          Originally posted by Joseph Coveney View Post
          I had already checked that and ruled it out as the cause of the discrepancy before posting. Results are quite a bit different from what is shown in the paper's Table 8. See below. You got something different using SAS yourself from what they report there?
          You are right, my mistake. I didn't compare the rest of the model. Thanks for showing me the same model using GSEM, which I am not very familiar with.

          Comment


          • #6
            I recommend looking into -gsem- if you're going to be doing more of this kind of thing. It's hard to beat for these situations.

            Comment


            • #7
              Thank you for the advice and guidance, Joseph. This is very helpful.

              Comment

              Working...
              X