1. Sumfreq / Sumtables (epidata-list@lists.umanitoba.ca)
Has anyone an idea on how to replace Sumfreq (or sumtables) with Epidata analysis? Thanks! Gilles / Paris
I understand the sumfreq and sumtables command to be able to do a frequency or crosstable but instead of counting observations the counting would be made with another variable.
Example from testdata: read bromar tab agegrp . freq agegrp agegrp Age (10 year intervals) No. % 10 35 0.92 20 489 12.92 30 1198 31.64 40 1337 35.31 50 642 16.96 60 75 1.98 70 9 0.24 80 1 0.03 Total 3786 100%
We can now collapse or aggregate the data file like this on agegrp: . agg agegrp /close //"agg" is the same as "aggregate" AGEGRP N 10 35 20 489 30 1198 40 1337 50 642 60 75 70 9 80 1
Now if we do a frequency on agegrp: agegrp Age (10 year intervals) No. % 10 1 12.50 20 1 12.50 30 1 12.50 40 1 12.50 50 1 12.50 60 1 12.50 70 1 12.50 80 1 12.50 Total 8 100%
We have one observation for each value of agegrp. 8 records or observations in total. By listing all observations we can see that the variable N tells us how many are in each agegrp: list AGEGRP N 10 35 20 489 30 1198 40 1337 50 642 60 75 70 9 80 1
So in principle we would like a "backwards" aggregate where the weight or counting is by "N":: . freq agegrp /weight=n /
Age (10 | year | intervals) | Freq. Percent ------------+-------------------- 10 | 35 0.92 20 | 489 12.92 30 | 1198 31.64 40 | 1337 35.31 50 | 642 16.96 60 | 75 1.98 70 | 9 0.24 80 | 1 0.03 ------------+------------------ Total | 3,786 100.00
Unfortunately the table just shown was made in Stata not in EpiData Analysis, but I have added sumfreq and sumtables to the workplan.
Please discuss on the list if the /weight= x principle is what we want. I have met various needs for weighting, but assume that the simple frequency weight shown here is sufficient.
Regards
Jens Lauritsen EpiData Association
Hi,
The command " freq agegrp /weight=n " would be sufficient to replace sumfreq. These weight variable will rarely be defined as an exposure or outcome variable but will mostly be like number, count, cases, occurence etc... Then we will use it more likely under the form : freq disease /weight=count and tables agegroup disease /weight=count
This will be very usefull for surveillance system where data are collected not as individual record but as cluster of cases.
Thanks Gilles
epidata-list@lists.umanitoba.ca a écrit :
- Sumfreq / Sumtables (epidata-list@lists.umanitoba.ca)
Has anyone an idea on how to replace Sumfreq (or sumtables) with Epidata analysis? Thanks! Gilles / Paris
I understand the sumfreq and sumtables command to be able to do a frequency or crosstable but instead of counting observations the counting would be made with another variable.
Example from testdata: read bromar tab agegrp . freq agegrp agegrp Age (10 year intervals) No. % 10 35 0.92 20 489 12.92 30 1198 31.64 40 1337 35.31 50 642 16.96 60 75 1.98 70 9 0.24 80 1 0.03 Total 3786 100%
We can now collapse or aggregate the data file like this on agegrp: . agg agegrp /close //"agg" is the same as "aggregate" AGEGRP N 10 35 20 489 30 1198 40 1337 50 642 60 75 70 9 80 1
Now if we do a frequency on agegrp: agegrp Age (10 year intervals) No. % 10 1 12.50 20 1 12.50 30 1 12.50 40 1 12.50 50 1 12.50 60 1 12.50 70 1 12.50 80 1 12.50 Total 8 100%
We have one observation for each value of agegrp. 8 records or observations in total. By listing all observations we can see that the variable N tells us how many are in each agegrp: list AGEGRP N 10 35 20 489 30 1198 40 1337 50 642 60 75 70 9 80 1
So in principle we would like a "backwards" aggregate where the weight or counting is by "N":: . freq agegrp /weight=n /
Age (10 | year | intervals) | Freq. Percent ------------+-------------------- 10 | 35 0.92 20 | 489 12.92 30 | 1198 31.64 40 | 1337 35.31 50 | 642 16.96 60 | 75 1.98 70 | 9 0.24 80 | 1 0.03 ------------+------------------ Total | 3,786 100.00
Unfortunately the table just shown was made in Stata not in EpiData Analysis, but I have added sumfreq and sumtables to the workplan.
Please discuss on the list if the /weight= x principle is what we want. I have met various needs for weighting, but assume that the simple frequency weight shown here is sufficient.
Regards
Jens Lauritsen EpiData Association
EpiData-list mailing list EpiData-list@lists.umanitoba.ca http://lists.umanitoba.ca/mailman/listinfo/epidata-list
participants (1)
-
epidata-list＠lists.umanitoba.ca