There is a kind of data I run into fairly often
which I have never known how to represent in R,
and nothing I've tried really satisfies me.
Consider for example
...
- injuries
...
- injuries to limbs
...
- injuries to extremities
...
- injuries to hands
- injuries to dominant hand
- injuries to non-dominant hand
...
...
...
This isn't ordinal data, because there is no
"left to right" order on the values. But there
IS a "part/whole" order, which an analysis should
respect, so it's not pure nominal data either.
As one particular example, if I want to
tabulate data like this, an occurrence of one
value should be counted as an occurrence of
*every* superordinate value.
Examples of such data include "why is this patient
being treated", "what drug is this patient being
treated with", "what geographic region is this
school from", "what biological group does this
insect belong to".
So what is the recommended way to represent
and the recommended way to analyse such data in R?
[[alternative HTML version deleted]]
Really this depends on the analysis you want to perform. In the past, I have used a super/sub two-column format as a compact, non-redundant representation for data entry, and after applying a recursive algorithm to convert this to a super/sub/level/id table where _all_ sub components have (duplicative) entries corresponding to each super component. But there is always the recursive list structure that formats such as yaml and json functions typically return. On May 29, 2022 9:54:44 PM PDT, Richard O'Keefe <raoknz at gmail.com> wrote:>There is a kind of data I run into fairly often >which I have never known how to represent in R, >and nothing I've tried really satisfies me. > >Consider for example > ... > - injuries > ... > - injuries to limbs > ... > - injuries to extremities > ... > - injuries to hands > - injuries to dominant hand > - injuries to non-dominant hand > ... > ... > ... > >This isn't ordinal data, because there is no >"left to right" order on the values. But there >IS a "part/whole" order, which an analysis should >respect, so it's not pure nominal data either. > >As one particular example, if I want to >tabulate data like this, an occurrence of one >value should be counted as an occurrence of >*every* superordinate value. > >Examples of such data include "why is this patient >being treated", "what drug is this patient being >treated with", "what geographic region is this >school from", "what biological group does this >insect belong to". > >So what is the recommended way to represent >and the recommended way to analyse such data in R? > > [[alternative HTML version deleted]] > >______________________________________________ >R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see >https://stat.ethz.ch/mailman/listinfo/r-help >PLEASE do read the posting guide http://www.R-project.org/posting-guide.html >and provide commented, minimal, self-contained, reproducible code.-- Sent from my phone. Please excuse my brevity.
The most common way of handling this, from what I have seen, and what I am also using myself, is to have multiple columns; one for each level of the tree. So in your example you would have (at least) five columns. A records with an injury to their dominant hand would have the following values in these columns: level0: injuries level1: injuries to limbs level2: injuries to extremities level3: injuries to hands level4: injuries to dominant hand That way you can easily aggregate to each of the levels. This solution becomes slightly more difficult if not each branch of the tree has the same depth. You can use NA's to fill non-existing levels. Often these types of variables come with codes that have a nesting structure. For example, in the regional codes we use, the first 4 digits are the municipality, the next two the city district and the next two the neighbourhood. So neighbourhoods have 6 digit codes. If I want to aggregate to municipality I can take the first 4 digits (substr). That way I only need one column and can calculate the others when needed. However, aggregating to municipality is so common that I often end up with a separate column anyway. In a hackaton at the uros-conference a while back, we did try to make a new type for this (https://github.com/uRosConf/categorical) but this never got completed enough to push to CRAN. HTH, Jan On 30-05-2022 06:54, Richard O'Keefe wrote:> There is a kind of data I run into fairly often > which I have never known how to represent in R, > and nothing I've tried really satisfies me. > > Consider for example > ... > - injuries > ... > - injuries to limbs > ... > - injuries to extremities > ... > - injuries to hands > - injuries to dominant hand > - injuries to non-dominant hand > ... > ... > ... > > This isn't ordinal data, because there is no > "left to right" order on the values. But there > IS a "part/whole" order, which an analysis should > respect, so it's not pure nominal data either. > > As one particular example, if I want to > tabulate data like this, an occurrence of one > value should be counted as an occurrence of > *every* superordinate value. > > Examples of such data include "why is this patient > being treated", "what drug is this patient being > treated with", "what geographic region is this > school from", "what biological group does this > insect belong to". > > So what is the recommended way to represent > and the recommended way to analyse such data in R? > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.
Hi Richard, Some years ago I had a try at illustrating Multiple Causes of Death (MCoD) data. I settled on what is sometimes called a "sizetree". You can see some examples in the sizetree function help page in "plotrix". Unfortunately I can't use the original data as it was confidential. Jim On Mon, May 30, 2022 at 2:55 PM Richard O'Keefe <raoknz at gmail.com> wrote:> > There is a kind of data I run into fairly often > which I have never known how to represent in R, > and nothing I've tried really satisfies me. > > Consider for example > ... > - injuries > ... > - injuries to limbs > ... > - injuries to extremities > ... > - injuries to hands > - injuries to dominant hand > - injuries to non-dominant hand > ... > ... > ... > > This isn't ordinal data, because there is no > "left to right" order on the values. But there > IS a "part/whole" order, which an analysis should > respect, so it's not pure nominal data either. > > As one particular example, if I want to > tabulate data like this, an occurrence of one > value should be counted as an occurrence of > *every* superordinate value. > > Examples of such data include "why is this patient > being treated", "what drug is this patient being > treated with", "what geographic region is this > school from", "what biological group does this > insect belong to". > > So what is the recommended way to represent > and the recommended way to analyse such data in R? > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.