Re: SPSS and replication/triplication
Bruce Weaver
> I am having a few issues with how to enter and analyze my data in SPSS.
>
> For simplicity, I'll just use an example where I'm trying to compare
> two groups, A and B. Let's say that both groups have 5 blood samples
> (n=5) taken from humans. I measure the glucose levels in each blood
> sample in triplicate, meaning that I take the 1st blood sample from
> group A and measure the glucose levels 3 times, and so on and so forth.
>
> So my question is how do i analyze this in SPSS? So i've run it where
> i have a "Group" column with 30 entries (half is 1 and half is 2, to
> denote A and B, respectively), and then in the "Glucose" column, i'll
> enter in the 15 glucose measurements from each group. However,
> comparing the means from this seems to indicate that n=15. I'm
> wondering how to tell SPSS that the 15 measurements are really coming
> from n=5 measured in triplicate?
>
> I've been reading around and it seems that there were suggestions that
> the replication should be added as a column, but i'm unsure. If there
> are any references or advice, i would greatly appreciate it. Thanks.
>
Questions like this should be directed to the SPSS group--I've
cross-posted this reply there.
From what you've said, I gather you want to perform an analysis that
includes one or more repeated measures factors, is that right? The
usual method of doing repeated measures ANOVA (Analyze -> GLM ->
Repeated Measures) requires one row per ID with multiple columns for the
repeated measures. So for your data, the column headings would look
something like this:
ID Group S1M1 S1M2 S1M3 S2M1 S2M2 S2M3 ... S5M1 S5M2 S5M3
I assume you want a mixed design ANOVA with Group as the between-Ss
factor. It is less clear whether you want two repeated measures
factors, Sample and Measure, or just one, Sample. If the latter, you
could compute the mean of the 3 measures for each sample, and use those
means as the 5 repeated measures.
--
Bruce Weaver
bwe...@lakeheadu.ca
www.angelfire.com/wv/bwhomedir
inva...@gmail.com
> inva...@gmail.com wrote:
> > I am having a few issues with how to enter and analyze my data in SPSS.
> >
> > For simplicity, I'll just use an example where I'm trying to compare
> > two groups, A and B. Let's say that both groups have 5 blood samples
> > (n=5) taken from humans. I measure the glucose levels in each blood
> > sample in triplicate, meaning that I take the 1st blood sample from
> > group A and measure the glucose levels 3 times, and so on and so forth.
> >
> > So my question is how do i analyze this in SPSS? So i've run it where
> > i have a "Group" column with 30 entries (half is 1 and half is 2, to
> > denote A and B, respectively), and then in the "Glucose" column, i'll
> > enter in the 15 glucose measurements from each group. However,
> > comparing the means from this seems to indicate that n=15. I'm
> > wondering how to tell SPSS that the 15 measurements are really coming
> > from n=5 measured in triplicate?
> >
> > I've been reading around and it seems that there were suggestions that
> > the replication should be added as a column, but i'm unsure. If there
> > are any references or advice, i would greatly appreciate it. Thanks.
> >
>
> Questions like this should be directed to the SPSS group--I've
> cross-posted this reply there.
>
> From what you've said, I gather you want to perform an analysis that
> includes one or more repeated measures factors, is that right?
Bruce, I don't believe that what I'm asking is a repeated measures. In
the example, what i'm saying is that I have one blood sample and from
that blood sample, I will measure the glucose level in triplicate (to
get an idea of the variation in let's say pipetting) as opposed to
measuring it at three distince times.
So for example, my data would look like something like this.
Group A
Sample 1 2.34 2.48 2.68
Sample 2 3.1 3.21 3.4
Sample 3 2.1 2.11 2.32
Sample 4 2.9 2.5 2.6
Sample 5 3.1 3.5 3.2
and then something similar for
Group B
Sample 1 12.34 12.48 12.68
Sample 2 13.1 13.21 13.4
Sample 3 12.1 12.11 12.32
Sample 4 12.9 12.5 12.6
Sample 5 13.1 13.5 13.2
So this is n=5. I usually take the average of all 15 numbers and the
standard deviation from all 15 measurements for each group. But I'm
unsure of how to enter this data in SPSS, because of the triplicate
measurements.
Bruce Weaver
So there is no one way in which to logically order the 3 measurements.
Why 3 measurements then? Are you just guarding against the possibility
of an off the wall result messing things up? If so, I assume you just
want the mean of the 3 readings.
>
> So for example, my data would look like something like this.
>
> Group A
> Sample 1 2.34 2.48 2.68
> Sample 2 3.1 3.21 3.4
> Sample 3 2.1 2.11 2.32
> Sample 4 2.9 2.5 2.6
> Sample 5 3.1 3.5 3.2
>
> and then something similar for
>
> Group B
> Sample 1 12.34 12.48 12.68
> Sample 2 13.1 13.21 13.4
> Sample 3 12.1 12.11 12.32
> Sample 4 12.9 12.5 12.6
> Sample 5 13.1 13.5 13.2
>
> So this is n=5. I usually take the average of all 15 numbers and the
> standard deviation from all 15 measurements for each group. But I'm
> unsure of how to enter this data in SPSS, because of the triplicate
> measurements.
I have a few questions.
1. Is each "sample" from a different person?
2. If YES to question 1, are the 5 people in Group B different people?
3. Are you happy with using the mean of the 3 measurements?
If "sample" means person, and groups A and B are two independent groups
(i.e., different people), and if you are happy with the mean of each
sample, then it sounds like you want an independent groups t-test. To
do it, set up your data like this:
Group G1 G2 G3
1 2.34 2.48 2.68
1 3.1 3.21 3.4
1 2.1 2.11 2.32
1 2.9 2.5 2.6
1 3.1 3.5 3.2
2 12.34 12.48 12.68
2 13.1 13.21 13.4
2 12.1 12.11 12.32
2 12.9 12.5 12.6
2 13.1 13.5 13.2
Then compute the mean glucose level for each sample and perform the
t-test like this:
compute meangluc = mean(g1,g2,g3).
exe.
T-TEST
GROUPS = group(1 2) /
VARIABLES = meangluc .
inva...@gmail.com
1 and 2. Each sample is from a different person. The people in Group A
are different from Group B. From each person, a blood sample is taken.
From that blood sample, the triplicate measurement is made.
3. The reason for the triplicate measurement is to guard against some
of the wall reading, for example, a pipetting error.
>
> If "sample" means person, and groups A and B are two independent groups
> (i.e., different people), and if you are happy with the mean of each
> sample, then it sounds like you want an independent groups t-test. To
> do it, set up your data like this:
>
> Group G1 G2 G3
> 1 2.34 2.48 2.68
> 1 3.1 3.21 3.4
> 1 2.1 2.11 2.32
> 1 2.9 2.5 2.6
> 1 3.1 3.5 3.2
> 2 12.34 12.48 12.68
> 2 13.1 13.21 13.4
> 2 12.1 12.11 12.32
> 2 12.9 12.5 12.6
> 2 13.1 13.5 13.2
>
> Then compute the mean glucose level for each sample and perform the
> t-test like this:
>
> compute meangluc = mean(g1,g2,g3).
> exe.
>
> T-TEST
> GROUPS = group(1 2) /
> VARIABLES = meangluc .
>
>
The t-test example seems to make sense to me. Let's say I have 3
groups (Group A, Group B, Group C) and I want to do ANOVA, though. I
would enter the data as this? To me, this seems like a one-way ANOVA.
However, under GLM- univariate, I would only be able to enter one
dependent variable (either G1, G2, or G3). I'm not sure if this is a
repeated measures though. The blood samples taken from the different
groups are all taken at one, same timepoint. The triplicate
measurements are to ensure that there are no weird pipetting errors.
The average and standard deviation I want is from the total 15
measurements from each group. In this case, I would just have my group
column and a G column. However, if I do post-hoc tests, i'm afraid it
would take my n to be 15 instead of 5.
Group G1 G2 G3
1 2.34 2.48 2.68
1 3.1 3.21 3.4
1 2.1 2.11 2.32
1 2.9 2.5 2.6
1 3.1 3.5 3.2
2 12.34 12.48 12.68
2 13.1 13.21 13.4
2 12.1 12.11 12.32
2 12.9 12.5 12.6
2 13.1 13.5 13.2
3 23.1 23.5 23.8
3 27.4 27.1 27.2
3 22.5 22.1 22.1
3 24.6 25.7 24.1
3 25.3 2.5.3 25.1
vs.
Group G
1 2.34
1 3.1
1 2.1
1 2.9
1 3.1
1 2.68
1 3.4
1 2.32
1 2.6
1 3.2
1 2.48
1 3.21
1 2.11
1 2.5
1 3.5
Bruce Weaver
>>
>> Then compute the mean glucose level for each sample and perform the
>> t-test like this:
>>
>> compute meangluc = mean(g1,g2,g3).
>> exe.
>>
>> T-TEST
>> GROUPS = group(1 2) /
>> VARIABLES = meangluc .
>>
>>
>
> The t-test example seems to make sense to me. Let's say I have 3
> groups (Group A, Group B, Group C) and I want to do ANOVA, though. I
> would enter the data as this? To me, this seems like a one-way ANOVA.
> However, under GLM- univariate, I would only be able to enter one
> dependent variable (either G1, G2, or G3). I'm not sure if this is a
If you had 3 groups, you could compute MeanGluc just as above, but then,
instead of doing a t-test, you would do a one-way ANOVA. There are (at
least) 3 ways to do that in SPSS:
1. Analyze -> Compare Means -> Means
2. Analyze -> Compare Means -> One-way ANOVA
3. Analyze -> GLM -> Univariate
Here's an example of syntax for the 3rd method:
UNIANOVA
MeanGluc BY Group
/EMMEANS = TABLES(AgeGroup)
/CRITERIA = ALPHA(.05)
/DESIGN = Group .
You could add a /POSTHOC sub-command, if desired.
Ray Koopman
> [...]
> Bruce, I don't believe that what I'm asking is a repeated measures. In
> the example, what i'm saying is that I have one blood sample and from
> that blood sample, I will measure the glucose level in triplicate (to
> get an idea of the variation in let's say pipetting) as opposed to
> measuring it at three distince times.
>
> So for example, my data would look like something like this.
>
> Group A
> Sample 1 2.34 2.48 2.68
> Sample 2 3.1 3.21 3.4
> Sample 3 2.1 2.11 2.32
> Sample 4 2.9 2.5 2.6
> Sample 5 3.1 3.5 3.2
>
> and then something similar for
>
> Group B
> Sample 1 12.34 12.48 12.68
> Sample 2 13.1 13.21 13.4
> Sample 3 12.1 12.11 12.32
> Sample 4 12.9 12.5 12.6
> Sample 5 13.1 13.5 13.2
>
> So this is n=5. I usually take the average of all 15 numbers and the
> standard deviation from all 15 measurements for each group. But I'm
> unsure of how to enter this data in SPSS, because of the triplicate
> measurements.
Looks to me like a hierarchical design:
Groups, Samples|Group, and Replications|Sample|Group,
with Groups fixed and Samples & Replications random.
The error term for G is S|G. The error term for S|G is R|S|G.
My quick-and-dirty non-SPSS results (which should be checked!)
for the data you gave:
Source df SS MS F p
G 1 750.0000 750.0000 1111.02 .716e-9
S|G 8 5.4005 0.6751 21.87 .275e-7
R|S|G 20 0.6173 0.0309
Total 29 756.0178
Erkki.Ko...@helsinki.fi.invalid
: From that blood sample, the triplicate measurement is made.
I would suggest you to use SPSS Linear Mixed Model. Like:
MIXED result
BY factor1 factor2
/METHOD = REML
/PRINT = SOLUTION TESTCOV
/FIXED = factor1 factor2 | SSTYPE(3)
/RANDOM = INTERCEPT | SUBJECT(person) COVTYPE(UN) .
This takes into accout possible depence through person (blood sample
with three measurements, and lets you to see and take into account
possible source of variation within this sample/person).
See Norusis, Marija: SPSS 14.0 Advanced Statistical procedures
Companion, Chapter 10 (Linear mixed models).
HTH, Cheers, Erkki
040-5024491 <http://www.helsinki.fi/people/Erkki.Komulainen/>
Erkki.Ko...@helsinki.fi.invalid
:MIXED result
: BY factor1 factor2
: /METHOD = REML
: /PRINT = SOLUTION TESTCOV
: /FIXED = factor1 factor2 | SSTYPE(3)
: /RANDOM = INTERCEPT | SUBJECT(person) COVTYPE(UN) .
:This takes into accout possible depence through person (blood sample
:with three measurements, and lets you to see and take into account
:possible source of variation within this sample/person).
:See Norusis, Marija: SPSS 14.0 Advanced Statistical procedures
:Companion, Chapter 10 (Linear mixed models).
I forgot an important thing. SPSS Mixed Module requires the data to be
in so called _narrow form_ (compared to the wide form which is the usual
form). You have to have the "repeated measures" stacked. See the
hierarchical linear model (multilevel) litterature.
Erkki
:040-5024491 <http://www.helsinki.fi/people/Erkki.Komulainen/>
040-5024491 <http://www.helsinki.fi/people/Erkki.Komulainen/>