Example L5 Ordered response model

 

In this example we analyse ordinal data on changes in the severity of Schizophrenia in data from the National Institute of Mental Health (NIMH), see Rabe-Hesketh and Skrondal (2005, Ch 5). Patients were randomly allocated to 1 of 4 treatments (Placebo, Chlorpromazine, Fluphenazine, Thioridazine) and examined weekly for up to 6 weeks. The data have previously been analysed by Hedeker and Gibbons (1996a), Gibbons et al. (1998) and Gibbons and Hedeker (1994). In this example we follow Rabe-Hesketh and Skrondal (2005, Ch 5) and combine the treatments Chlorpromazine, Fluphenazine and Thioridazine.

 

 

Data description

 

Number of observations (rows): 1603

Number of variables (columns): 8

 

Variables:

 

id=subject identifier

week=week (0,1,2,…,6) of assessment since randomisation to treatment;

imps=score on item 79 of the Inpatient Multidimensional Psychiatric Scale (IMPS) of Lorr and Klextt (1966).

treatment=1 if subject was in the treatment group (drug), 0 otherwise;

 

Following Rabe-Hesketh and Skrondal (2005, Ch 5) and Hedeker and Gibbons (1996a) we recode the imps variable into impso, an ordinal variable with 4 categories, i.e.

 

impso=1 if 1<=imps<=2.4;

impso=2 if 2.5<=imps<4.4

impso=3 if 4.5<=imps<=5.4

impso=4 if 5.5<=imps<=7

 

Following Rabe-Hesketh and Skrondal (2005, Ch 5) we also use the square root of week (weeksqrt) instead of week in its raw form as an explanatory variable. Similarly we also produce a new variable from the interaction (interact) of weeksqrt and treatment.  

 

 

References

 

Gibbons, R. D.,  Hedeker, C., Waterneaux, C., & Davis, J.M., (1988), Random regression models: A comprehensive approach to the analysis of longitudinal psychiatric data, Psychopharmacology Bulletin, 24, 438-443

Gibbons, R. D., & Hedeker, C., (1994), Application of random effects probit regression models, Journal of Consulting and Clinical Psyschology.

Hedeker, D., & Gibbons, R.D., (1996), Applied longitudinal data anlaysis, Chichester, UK, Wiley

Rabe-Hesketh, S., and Skrondal, A., (2005), Multilevel and Longitudinal Modelling using Stata, Stata Press, Stata Corp, College Station, Texas.

 

 

 

 

The first few lines of schiz.dat look like

 

 

 

Sabre commands

 

out schiz.log

trace schiz.trace

data id imps week treatment sex weeksqrt interact impso

read schiz.dat

case id

yvar impso

ordered y

lfit weeksqrt treatment interact

dis m

dis e

fit weeksqrt treatment interact

dis m

dis e

stop

 

 

 

Sabre log file

 

<S> trace schiz.trace

<S> data id imps week treatment sex weeksqrt interact impso

<S> read schiz.dat

 

       1603 observations in dataset

 

<S> case id

<S> yvar impso

<S> ordered y

<S> lfit weeksqrt treatment interact

 

    Iteration       Log. lik.         Step      End-points     Orthogonality

                                     length    0          1      criterion

    ________________________________________________________________________

        1          -2239.6731        1.0000    fixed  fixed       18.971

        2          -1947.5622        0.5000    fixed  fixed       46.248

        3          -1885.7509        1.0000    fixed  fixed       131.39

        4          -1879.9699        1.0000    fixed  fixed       248.84

        5          -1878.6522        1.0000    fixed  fixed       1158.2

        6          -1878.2740        1.0000    fixed  fixed       253.33

        7          -1878.0969        1.0000    fixed  fixed       54.836

        8          -1878.0969        1.0000    fixed  fixed

 

<S> dis m

 

    X-vars            Y-var

    ______________________________

    weeksqrt          impso

    treatment

    interact

 

    Univariate model

    Standard ordered logit

 

    Number of observations             =    1603

 

    X-var df           =     3

 

    Log likelihood =     -1878.0969     on    1600 residual degrees of freedom

 

<S> dis e

 

    Parameter              Estimate         Std. Err.

    ___________________________________________________

    weeksqrt              -0.53665          0.11081

    treatment             -0.60428E-03      0.18833

    interact              -0.75097          0.12768

    cut1                   -3.8073          0.18986

    cut2                   -1.7602          0.17027

    cut3                  -0.42211          0.16363

 

<S> fit weeksqrt treatment interact

 

    Initial Homogeneous Fit:

 

    Iteration       Log. lik.         Step      End-points     Orthogonality

                                     length    0          1      criterion

    ________________________________________________________________________

        1          -2239.6731        1.0000    fixed  fixed       18.971

        2          -1947.5622        0.5000    fixed  fixed       46.248

        3          -1885.7509        1.0000    fixed  fixed       131.39

        4          -1879.9699        1.0000    fixed  fixed       248.84

        5          -1878.6522        1.0000    fixed  fixed       1158.2

        6          -1878.2740        1.0000    fixed  fixed       253.33

        7          -1878.0969        1.0000    fixed  fixed       54.836

        8          -1878.0969        1.0000    fixed  fixed

 

 

    Iteration       Log. lik.         Step      End-points     Orthogonality

                                     length    0          1      criterion

    ________________________________________________________________________

        1          -1775.2726        1.0000    fixed  fixed       14.707

        2          -1717.2806        1.0000    fixed  fixed       12.350

        3          -1703.2212        1.0000    fixed  fixed       119.24

        4          -1701.5195        1.0000    fixed  fixed       340.03

        5          -1701.2283        1.0000    fixed  fixed       337.78

        6          -1701.1745        1.0000    fixed  fixed       211.94

        7          -1701.1622        1.0000    fixed  fixed       18.610

        8          -1701.1622        1.0000    fixed  fixed

 

<S> dis m

 

    X-vars            Y-var             Case-var

    ________________________________________________

    weeksqrt          impso             id

    treatment

    interact

 

    Univariate model

    Standard ordered logit

    Gaussian random effects

 

    Number of observations             =    1603

    Number of cases                    =     437

 

    X-var df           =     3

    Scale df           =     1

 

    Log likelihood =     -1701.1622     on    1599 residual degrees of freedom

 

<S> dis e

 

    Parameter              Estimate         Std. Err.

    ___________________________________________________

    weeksqrt              -0.76775          0.13107

    treatment             -0.81896E-01      0.31086

    interact               -1.2069          0.15280

    cut1                   -5.8917          0.33345

    cut2                   -2.8572          0.28969

    cut3                  -0.73532          0.27343

    scale                   1.9520          0.12113

 

<S> stop