|
160. Total time spent talking |
|
TTALK |
NGROUP
The following are the labels for NGROUP.
value labels ngroup
1 'high school'
2 'graduate students1'
3 'graduate
students2'
4 'general'
5 'Psychiatri Pt.'.
The
following syntax file is generated by the following clicks:
Click
Analyze
Click
General Linear Model
Click
Multivariate
Select
dependent variables
Click
right delta for Dependent Variables
Select
ngroup
Click
right delta for Fixed Factors
Click
Options
Select
ngroup
Click
right delta
Select
Compare Main Effects
Select
Descriptive Statistics
Click
Contine
Click
Paste
The above clicks produce the following
syntax file:
|
Syntax file name = lsyman6a |
|
*get file =
"f:\rdda\lsqrnd5.sav". GLM
enjoy fgood worth fearful angry tense shy wornout fitin BY ngroup
/METHOD = SSTYPE(3)
/INTERCEPT = INCLUDE
/EMMEANS = TABLES(OVERALL)
/EMMEANS = TABLES(ngroup) COMPARE ADJ(LSD)
/PRINT = DESCRIPTIVE
/CRITERIA = ALPHA(.05)
/DESIGN = ngroup . |
The output follows:
General
Linear Model
Estimated
Marginal Means
2.
NGROUP
What does it mean that the MANOVA is
significant? All of the y variables are
combined into a single dimension very much like a factor in factor analysis, so
that each has a weight (like a beta weight) and are then combined to form a
single variable. The score is like our
old Y' (Y primed) formula. So that, in
fact, an ANOVA is computed using this new Y' and the Y variables. Consequently, the significant MANOVA
indicates that there is a difference between at least two of the groups when
comparing the means on this new Y variable that is due to chance at less than
.0001 level of probability.
Consequently, there is a lot that it does not indicate: it does not
indicate which two groups might be different, nor does it indicate which of the
combined variables might be producing that effect. It is therefore necessary to determine which
variables are producing the effect and which groups are producing the effect.
It may be useful at this point to
identify the advantages and disadvantages of using MANOVA. The advantages are: (1) it gives you an
overall indication of the effect of multiple y variables (dependent variables)
that you could not get using ANOVA, for example, (2) it guards against Type I
errors in that it gives an estimate for all of the variables at once (if many
ANOVAs were computed each one individually gives a biased estimate depending on
the number computed), and (3) under certain conditions it may reveal
differences not shown in separate ANOVAs.
The disadvantages are: (1) the ambiguity of interpretation, (2) the
situations where MANOVA is more powerful is quite limited, most often is it
less powerful, (3) ] must eventually be computed to assess the meaning of the
result and further pairwise comparisons must eventually be computed to assess
which groups produce the effects. My
recommendation is to use MANOVA with caution.
In order to show the effects the MANOVA
will be computed again without one of the groups.
|
File name = laqman11a.sps |
|
*get file =
"f:\rdda\lsqrnd5.sav". GLM
enjoy fgood worth fearful angry tense shy wornout fitin BY ngroupm
/METHOD = SSTYPE(3)
/INTERCEPT = INCLUDE
/EMMEANS = TABLES(OVERALL)
/EMMEANS = TABLES(ngroupm) COMPARE ADJ(LSD)
/PRINT = DESCRIPTIVE
/CRITERIA = ALPHA(.05)
/DESIGN = ngroupm . |
General
Linear Model
The only changes between this jobstreat and the previous os
(lsqman6a.sps) is that this jobstream does not include the group os psychaitric
patients.
Notice
the overall MANOVA is not significant and only two of the ANOVAs are
significant (SAY and STANDUP). This
shows what effect a single group can have on the results. In the next run only the two y variables will
be use that were significant at the univariate level. However, it should be noted that the tests of
significance are misleading because of the large number of ANOVAs that have
been computed using the Bonferroni adjustment:
Using
the .01 level of significance for the above problem the actual level would be:
alpha=1‑(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)(1‑.01)
alpha
= .122. The .009 alpha becomes .111.
Note that the MANOVA is significant as well as the two univariate tests. When a number of DVs are significant it is necessary to perform a "step‑down" analysis in order to determine whether a single variable may account for the significance in very much the same way that in multiple regression it is necessary to determine if there is a single variable that accounts for all of the variance in y. The method of doing this is compute a number of DVs identifying the theoretically most important variable or the empirically highest loading variable as the DV and all other y variables as covariates. Any variable, when identified as the DV when the ANCOVA remains significant is considered a significant variable. The next job stream treats the most empirically strong variable as the DV and the weakest variable as a covariate in an ANCOVA analysis.
Notice
that the alpha is significant at the .006 level indicating that SAY is
significant when variance for STANDUP is taken out removed. This is the same procedure that is used when
the variance of one variable is removed from the multiple regression formula
and the remaining variable is tested for significance. Next STANDUP is tested when SAY is removed.