UNIANOVA with a random factor

[Hongmei-2]

Dear all,

I met a problem while using UNIANOVA with a random factor. The data were
collected among urban pregnant women using cluster sampling in 2004 &
2006. The sample size seems to be adequate (n=574). The distribution of
dependent variable and all independent variables seems to be fine too. But
I could not get the estimated marginal means. I don't know what the reason
is. I am attaching the syntax and selected outputs below FYI. I appreciate
you kind help. Thanks in advance!

Syntax is pasted below.

*note: dv: dependent variable;
       year: study year, with values 2004 & 2006;
       psu: primary sampling unit, about 70 units with number of
interviewee varing from 2 to 13;
       ses: social-economic status*********.

UNIANOVA
dv BY year psu WITH age ses
/RANDOM=psu
/METHOD=SSTYPE(3)
/INTERCEPT=INCLUDE
/EMMEANS=TABLES(year) with (age=MEAN ses=MEAN) COMPARE ADJ(LSD)
/CRITERIA=ALPHA(.05)
/DESIGN=year psu age ses.


Selective outputs are pasted below:

 Tests of Between-Subjects Effects
Dependent Variable: dv  dependent variable

Source                 Type III Sum             Mean
                        of Squares df Square   F Sig.
Intercept  Hypothesis    75.172         1 75.172 238.997 .000
     Error         56.509         179.661 .315(a)
year    Hypothesis    .000         0 . . .
     Error           .         . .(b)
psu    Hypothesis  40.945         77 .532 2.913 .000
     Error         90.005         493 .183(c)
age    Hypothesis    .079         1 .079 .430 .512
     Error         90.005         493 .183(c)
ses    Hypothesis    .259         1 .259 1.417 .234
     Error         90.005       493 .183(c)

a  .378 MS(psu) + .622 MS(Error)
b  Cannot compute the appropriate error term using Satterthwaite's method.
c   MS(Error)

Estimated marginal means are not estimable.

[Marta García-Granero]

Hi Hongmei

I'm not really sure, because you don't give details enough of the
design, but I think that "psu" (primary sampling unit) must be nested
inside "year". Also, if the design is unbalanced, then type I sums of
squares are a better choice. Try this modified syntax:

UNIANOVA
dv BY year psu WITH age ses
 /RANDOM=psu
 /METHOD=SSTYPE(1)
 /INTERCEPT=INCLUDE
 /EMMEANS=TABLES(year) with (age=MEAN ses=MEAN) COMPARE ADJ(LSD)
 /CRITERIA=ALPHA(.05)
 /DESIGN=year psu(year) age ses.

Anyway, with UNIANOVA the comparisons of the EMMEANS will be flawed
(the wrong error term is used). Try using MIXED.

H> I met a problem while using UNIANOVA with a random factor. The data were
H> collected among urban pregnant women using cluster sampling in 2004 &
H> 2006. The sample size seems to be adequate (n=574). The distribution of
H> dependent variable and all independent variables seems to be fine too. But
H> I could not get the estimated marginal means. I don't know what the reason
H> is. I am attaching the syntax and selected outputs below FYI. I appreciate
H> you kind help. Thanks in advance!

H> Syntax is pasted below.

H> *note: dv: dependent variable;
H>        year: study year, with values 2004 & 2006;
H>        psu: primary sampling unit, about 70 units with number of
H> interviewee varing from 2 to 13;
H>        ses: social-economic status*********.

H> UNIANOVA
H> dv BY year psu WITH age ses
H> /RANDOM=psu
H> /METHOD=SSTYPE(3)
H> /INTERCEPT=INCLUDE
H> /EMMEANS=TABLES(year) with (age=MEAN ses=MEAN) COMPARE ADJ(LSD)
H> /CRITERIA=ALPHA(.05)
H> /DESIGN=year psu age ses.


H> Selective outputs are pasted below:

H>  Tests of Between-Subjects Effects
H> Dependent Variable: dv  dependent variable

H> Source                 Type III Sum             Mean
H>                         of Squares df Square   F Sig.
H> Intercept  Hypothesis    75.172         1 75.172 238.997 .000
H>      Error         56.509         179.661 .315(a)
H> year    Hypothesis    .000         0 . . .
H>      Error           .         . .(b)
H> psu    Hypothesis  40.945         77 .532 2.913 .000
H>      Error         90.005         493 .183(c)
H> age    Hypothesis    .079         1 .079 .430 .512
H>      Error         90.005         493 .183(c)
H> ses    Hypothesis    .259         1 .259 1.417 .234
H>      Error         90.005       493 .183(c)

H> a  .378 MS(psu) + .622 MS(Error)
H> b  Cannot compute the appropriate error term using Satterthwaite's method.
H> c   MS(Error)

H> Estimated marginal means are not estimable.

--
Regards,
Dr. Marta García-Granero,PhD           mailto:[hidden email]
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)

[John den Dunnen]

Hello,

Maybe somebody can help me with my off topic question.

I calculated the standard deviation of 15 single measurements.

How can I estimate the standard deviation for duplo and triplo measurements.

With regards,

John