Help with Cochran Test please? CORRECTED VERSION!

[Charlotte-9]

Sorry everyone....ignore my last post!  This is a corrected version (got
some of the numbers mixed up when I pasted my table in the last post)....


Dear all,

I am trying to analyse uptake of cancer screening across three screening
rounds for common groups of patients (identified by age).  It was
suggested a while ago that I use the Cochran test for this since I have
dependent data and a binary outcome variable. However, after reading lots
if info about this test, I am still not entirely sure how to apply it.
Could someone give me some ideas, please?

The data table is as follows:

Age group    Number of people screened    Total people
                                                (in each age group)
           Round 1    Round 2   Round 3
1 (youngest)   149      153       146            199
2    5911      5908      5636           7274
3    6788      6792      6305           8288
4 (oldest)   5621      5569      5035       6976

To summarise, there are 22737 people in total and they were
either ‘screened’ or ‘not screened’ in each round.   Each age group (1-4)
contains the same people from round to round.  So, in other words, of the
199 people who make up age group 1, 149 of them were screened in round 1,
153 of these same people in round 2 and 146 of them in round 3 etc.

I basically want to see if screening behaviour changes for the 'same'
people as they go through the screening process.  I have already
discovered that age has a negative effect on screening uptake, but I just
want to carry out a concrete test to identify differences for these common
groups.

Thanks in advance for any thoughts/help!

Lou

[Marta García-Granero]

Hi Lou,

I remember your first post on the topic, and I also remember that my
answer was that to perform Cochran test for dependent samples, you
needed the original (non aggregated data) because the test analyzes
individual data (case by case).

Perhaps the following example dataset will show you what I'm trying to
say:

* Sample dataset  *.
DATA LIST FREE /march june august (3 F8.0).
BEGIN DATA
0 1 0 0 0 0 0 1 0 1 0 0 1 1 0 1 0 0
0 0 0 1 1 0 1 1 0 1 0 0 1 0 0 1 1 0
0 1 1 0 1 1 0 0 0 0 1 0 0 1 0 1 0 0
1 0 0 1 0 0 1 0 1 1 1 1 1 0 0 1 1 0
END DATA.
VALUE LABEL march TO august 0'Negative' 1'Positive'.
LIST.

This data show the evolution of 24 Acacia trees after a symbiotic
insect is eliminated. Data are clearly related because every tree is
analyzed 3 times (march/june august).

NPAR TESTS /COCHRAN = march june august.

Since you have aggregated data, you've lost the information concerning
matching of data, and you have to treat them as independent: use
linear-by-linear association test.

Regards

Marta

L> I am trying to analyse uptake of cancer screening across three screening
L> rounds for common groups of patients (identified by age).  It was
L> suggested a while ago that I use the Cochran test for this since I have
L> dependent data and a binary outcome variable. However, after reading lots
L> if info about this test, I am still not entirely sure how to apply it.
L> Could someone give me some ideas, please?

L> The data table is as follows:

L> Age group    Number of people screened    Total people
L>                                                 (in each age group)
L>            Round 1    Round 2   Round 3
L> 1 (youngest)   149      153       146            199
L> 2    5911      5908      5636           7274
L> 3    6788      6792      6305           8288
L> 4 (oldest)   5621      5569      5035       6976

L> To summarise, there are 22737 people in total and they were
L> either ‘screened’ or ‘not screened’ in each round.   Each age group (1-4)
L> contains the same people from round to round.  So, in other words, of the
L> 199 people who make up age group 1, 149 of them were screened in round 1,
L> 153 of these same people in round 2 and 146 of them in round 3 etc.

L> I basically want to see if screening behaviour changes for the 'same'
L> people as they go through the screening process.  I have already
L> discovered that age has a negative effect on screening uptake, but I just
L> want to carry out a concrete test to identify differences for these common
L> groups.

[Charlotte-9]

Hi Marta (or anyone else who responds),

Thanks for your help with the Cochran Test.  I ran the test on my own data
and obtained the following:


Within age group 2, 5911 of the people were screened in rd1, 5908 in rd 2
and 5636 in rd 3.  I therefore did the following:

NPAR TESTS /COCHRAN = Screened_rd1 Screened_rd2 Screened_rd5.

(using the original screened variables which are coded 1 = screened, 0 =
not screened).

The results gave the following:
N = 7274
Cochran’s Q = 79.121*
df = 2
Asymp. Sig = 0.0001
* 0 is treated as a success.

I assume I have done everything right here from what I understand.  My
only question/query is that sometimes the test treats ‘1’ as success and
sometimes it treats ‘0’ as success.  Is there a way to always ensure
that ‘1’ is treated as success?  Why does the test do this in SPSS?  How
does this affect my interpretation of the results?

Kind regards,

Lou


On Fri, 30 Jun 2006 18:24:45 +0200, =?ISO-8859-1?B?
TWFydGEgR2FyY+1hLUdyYW5lcm8=?= <[hidden email]> wrote:

>Hi Lou,
>
>I remember your first post on the topic, and I also remember that my
>answer was that to perform Cochran test for dependent samples, you
>needed the original (non aggregated data) because the test analyzes
>individual data (case by case).
>
>Perhaps the following example dataset will show you what I'm trying to
>say:
>
>* Sample dataset  *.
>DATA LIST FREE /march june august (3 F8.0).
>BEGIN DATA
>0 1 0 0 0 0 0 1 0 1 0 0 1 1 0 1 0 0
>0 0 0 1 1 0 1 1 0 1 0 0 1 0 0 1 1 0
>0 1 1 0 1 1 0 0 0 0 1 0 0 1 0 1 0 0
>1 0 0 1 0 0 1 0 1 1 1 1 1 0 0 1 1 0
>END DATA.
>VALUE LABEL march TO august 0'Negative' 1'Positive'.
>LIST.
>
>This data show the evolution of 24 Acacia trees after a symbiotic
>insect is eliminated. Data are clearly related because every tree is
>analyzed 3 times (march/june august).
>
>NPAR TESTS /COCHRAN = march june august.
>
>Since you have aggregated data, you've lost the information concerning
>matching of data, and you have to treat them as independent: use
>linear-by-linear association test.
>
>Regards
>
>Marta
>
>L> I am trying to analyse uptake of cancer screening across three

screening
>L> rounds for common groups of patients (identified by age).  It was
>L> suggested a while ago that I use the Cochran test for this since I have
>L> dependent data and a binary outcome variable. However, after reading
lots

>L> if info about this test, I am still not entirely sure how to apply it.
>L> Could someone give me some ideas, please?
>
>L> The data table is as follows:
>
>L> Age group    Number of people screened    Total people
>L>                                                 (in each age group)
>L>            Round 1    Round 2   Round 3
>L> 1 (youngest)   149      153       146            199
>L> 2    5911      5908      5636           7274
>L> 3    6788      6792      6305           8288
>L> 4 (oldest)   5621      5569      5035       6976
>
>L> To summarise, there are 22737 people in total and they were
>L> either ‘screened’ or ‘not screened’ in each round.   Each age group (1-

4)
>L> contains the same people from round to round.  So, in other words, of
the
>L> 199 people who make up age group 1, 149 of them were screened in round
1,
>L> 153 of these same people in round 2 and 146 of them in round 3 etc.
>
>L> I basically want to see if screening behaviour changes for the 'same'
>L> people as they go through the screening process.  I have already
>L> discovered that age has a negative effect on screening uptake, but I
just
>L> want to carry out a concrete test to identify differences for these
common
>L> groups.

[Marta García-Granero]

Hi Lou,

I *think* that the first value encountered is treated as a success.
See the difference for this dataset:

DATA LIST FREE /march june august (3 F8.0).
BEGIN DATA
0 1 0 0 0 0 0 1 0 1 0 0 1 1 0 1 0 0
0 0 0 1 1 0 1 1 0 1 0 0 1 0 0 1 1 0
0 1 1 0 1 1 0 0 0 0 1 0 0 1 0 1 0 0
1 0 0 1 0 0 1 0 1 1 1 1 1 0 0 1 1 0
END DATA.
VALUE LABEL march TO august 0'Negative' 1'Positive'.
LIST.

NPAR TESTS /COCHRAN = march june august.

SORT CASES BY march(D).
NPAR TESTS /COCHRAN = march june august.

L> The results gave the following:
L> N = 7274
L> Cochran’s Q = 79.121*
L> df = 2
L> Asymp. Sig = 0.0001
L> * 0 is treated as a success.

L> I assume I have done everything right here from what I understand.  My
L> only question/query is that sometimes the test treats ‘1’ as success and
L> sometimes it treats ‘0’ as success.  Is there a way to always ensure
L> that ‘1’ is treated as success?  Why does the test do this in SPSS?  How
L> does this affect my interpretation of the results?

It doesn't affect at all.

BTW, do you want to run post-hoc comparisons (compare "rd" pairwise)?

Other topic: you mentioned that you wanted to test whether there were
differences in trend among age groups. These could be done (with a bit
of manual work or writing some matrix code) adapting the heterogeneity
test described by Sokal & Rohlf in their book "Biometry". Try to take
a look at it (I can't fotocopy pages of mine, because it is in
Spanish). I'm off soon until the 17 of july (I always abandon my city
when San Fermines, the "Running of the Bulls", because I can't stand
the sight of hundreds of drunkards sleeping in the middle of our
beautiful city gardens, treading the flowers, and I don't enjoy going
to the arena either to watch a splendid animal die either...). If you
can wait until I'm back, then I could write some syntax for that task.

Regards

Marta