Re: statalist-digest V4 #4961: Re: st: Value labels won't show on box plot axis

[Nick Cox <njcoxstata@gmail.com>]

Whether you can (validly) take means of ordinal scales (if that is
what we have here) is arguably not to the point. Boxplots show
medians, quartiles, etc. and not means. There's a judgement call on
quite how _useful_ those measures are with ordinal scales but they are
well defined.

I would usually prefer different plots myself  for ordinal scales but
(for once) was holding off on general pontification.
Nick
njcoxstata@gmail.com


On 2 August 2013 12:14, Allan Reese (Cefas) <allan.reese@cefas.co.uk> wrote:
> On 1 August 2013 10:46, Walsh, Lee <l.walsh@ucl.ac.uk> wrote:
>>> I am box plotting a variable grouped by another.
>>> I am telling stata to use the value labels for the y axis
>>> but it insists on showing the numerical values.
>>>
>>> graph box response, over(statementNum) ylabel(1(1)7, valuelabel)
>
> On 1 Aug 2013 11:00:00 +0100, Nick Cox <njcoxstata@gmail.com> agreed
>> This didn't work either in my experiments. Here's a replicable example
>> and a work-around:
>
>> . sysuse auto
>> . graph box foreign, over(rep78) yla(0 1, valuelabel)
>
> [and built a list of re-labels in a macro named -call-]
> ...
>> forval i = 1/7 {
>>       local call `call' `i' "`: label (response) `i''"
>> }
> [which was then used in the -graph- command]
>  . graph box response, over(statementNum) yla(`call')
> -------------------------------------------------------
>
> That's an elegant and general piece of programming, but ignores the semantic question of why you would use a boxplot on a non-metric variable.  I take it that Lee's responses are a 7-point opinion scale. There are differing views on whether it is valid to compare means of scales, given the numeric codes are arbitrary beyond the ordering.  Extracting the five-number summary and making a boxplot seems very doubtful to me. Nick's cars example plot shows the effect of applying boxplots to a discrete variable with few values and multiple coincidences.
>
> It may be quicker and less baffling for small numbers of re-labels to write the code explicitly.  I've recently had numerous graphs where a particular value represents the "limit of detection".  That is certainly worth drawing attention to in the graph rather than just in the caption.  I also prefer natural labels when plotting a logged number.  Bacterial counts usually follow a lognormal distribution.
>
> graph box logbact, over(groupvar) ylab(1.3 "20=LoD" 2 "100" 3 "1000" ...)
>
> Allan
>
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/