Repeated Measures and Chi-Square Alternative
dickin3
measures across 6 conditions. Individuals were to perform a balance test and
maintain their balance, in the event that they lost their balance I would
record that as a fall. I had three groups of subjects, each group performed
the same 6 conditions (which involved 3 trials for each condition). An
individual could sustain a fall on any trial of any condition.
So my questions in how do I analyze this. The data results in a 6 condition
X 3 group - frequency table of number of falls. I am attempting to see if
one group falls more often than the other groups. I know I can look at this
qualitatively, but I would like to perform some sort of statistical analysis
on this data. At first thought I figured I should use Chi Square, but
repeated measures violates the assumptions of the Chi-Square analysis.
Any help would be appreciated.
Clark Dickin
email {ou8...@hotmail.com}
Rich Ulrich
As a concession to the non-continuous aspect of the
counts of 'number of falls' for a person, I would look
for and report on all cases who never fell at all, or who
fell every time, 6x3 = 18 falls.
I would expect everyone else to fall in the range of
3-15 falls. Exceptions might need to be noted and
counted and reported. Were the exceptions
occurring on the trials that were, respectively, 'easy'
and 'difficult' among the others? - if there's something
weird going on, you want to be able to spot it.
Assuming no bad-data surprises:
Then I would do a repeated measures ANOVA where
each subject was measured with the count, 0 to 3, of
the number of falls in that condition, for each of 6
conditions.
--
Rich Ulrich, wpi...@pitt.edu
http://www.pitt.edu/~wpilib/index.html
MikeN
Why such a messy design. You could have measured time before fall! Then you
would have used parametric statistics.
I would suggest that you forget the repeated testing by coming up with a
single score for eact subject. Can you do this? Then we can talk.
Check www.stat-tutor.com It is free.
"dickin3" <dic...@cox.net> wrote in message
news:emKT8.58670$PW.4...@news2.central.cox.net...
jim clark
> "dickin3" <dic...@cox.net> wrote in message
> news:emKT8.58670$PW.4...@news2.central.cox.net...
> > I am looking at analyzing some frequency data that involved
> repeated > measures across 6 conditions. Individuals were to
> perform a balance test and > maintain their balance, in the
> event that they lost their balance I would > record that as a
> fall. I had three groups of subjects, each group performed >
> the same 6 conditions (which involved 3 trials for each
> condition). An > individual could sustain a fall on any trial
> of any condition.
> > So my questions in how do I analyze this. The data results
> in a 6 condition > X 3 group - frequency table of number of
> falls. I am attempting to see if > one group falls more often
> than the other groups. I know I can look at this >
> qualitatively, but I would like to perform some sort of
> statistical analysis > on this data. At first thought I
> figured I should use Chi Square, but > repeated measures
> violates the assumptions of the Chi-Square analysis.
Performing a mixed analysis of variance with number of falls as
the dependent variable (0-3). You can use either GLM or MANOVA
(in syntax). Specify condition as the within-subject factor, and
group as the between subject. Assuming f1 to f6 represent number
of falls in the 6 conditions, the manova command would be:
manova f1 f2 f3 f4 f5 f6 by group(1 3) /wsf = cond(6).
As a 6x3 factorial could be insensitive to the interaction, you
might want to think of meaningful contrasts for the condition and
group variables (polynomial, planned contrasts).
Best wishes
Jim
============================================================================
James M. Clark (204) 786-9757
Department of Psychology (204) 774-4134 Fax
University of Winnipeg 4L05D
Winnipeg, Manitoba R3B 2E9 cl...@uwinnipeg.ca
CANADA http://www.uwinnipeg.ca/~clark
============================================================================
Terry Reedy
> measures across 6 conditions. Individuals were to perform a balance
test and
> maintain their balance, in the event that they lost their balance I
would
> record that as a fall. I had three groups of subjects, each group
performed
> the same 6 conditions (which involved 3 trials for each condition).
An
> individual could sustain a fall on any trial of any condition.
>
> So my questions in how do I analyze this. The data results in a 6
condition
> X 3 group - frequency table of number of falls. I am attempting to
see if
> one group falls more often than the other groups.
For this particular question, you could reduce data for each person to
number of falls out of 18 trials and try simple 3-group analysis of
variance. I am assuming that you have enough in each group to make
this sensible and would pay some attention to whether distribution for
each group is roughly bell-shaped with approximately same spread. To
include condition in analysis, see others posts.
Terry J. Reedy
Thom
> number of falls out of 18 trials and try simple 3-group analysis of
> variance. I am assuming that you have enough in each group to make
> this sensible and would pay some attention to whether distribution for
> each group is roughly bell-shaped with approximately same spread. To
> include condition in analysis, see others posts.
>
> Terry J. Reedy
Setting aside time travel and changing ther design. I'd try what
Terry suggested first. Repeated measures ANOVA should peform pretty
well as long as the mean number of falls is around 0.5 (0.2-0.8 is
often cited). Before the popularity of logistic regression the arcsin
transformation was also widely used in these situations. There are
repeated measures "versions" of logistic regression*, but they might
be overkill if the ANOVA model is reasonable (standard
diagnostics/checks of assumptions apply).
* multilevel/heirarchical logistic regression is available in MlWin,
SAS, HLM etc. and will do the job (but there are probably other
options).
Thom