Re: st: change grouping in Stata -graph- command

[William Buchanan <william@williambuchanan.net>]

Using the -by()- option will produce multiple graph regions in the same image.  If the user wants things to appear that way, there are two different approaches that will yield equivalent results:

1.) Compute and store the summary statistics of interest for each group (I typically store the results in a matrix); -svmat- the matrix with the results;  use the same graph command with the - (asis) - option before the variable names.

2.) Reshape the data set from wide to long format (e.g., store the responses in a single column vector and use an additional variable to identify the survey item of interest), then use two - over() - options to get the intended aesthetic:

clear
set obs 100
g byte treat=rbinomial(1,.5) ///  Create a random treatment indicator
g loc = runiform() /// Generate a random uniform variable for location
sort loc /// Sort on the values of location
g price = runiform() /// Generate a random uniform variable for price
sort loc price ///  Sort on the values of location and price
g temp = runiform() /// Generate a random uniform variable for temperature
foreach v of var loc price temp { 
	egen `v'_b = cut(`v'), group(7) /// Create new versions of the variables with 7 discrete values
	drop `v'
}
rename (loc_b price_b temp_b)(v1 v2 v3) /// rename the variables to make it easier to reshape the data
g id = _n ///  Generate individual IDs
reshape long v, i(id treat) j(var) ///   Reshape the data and create a new variable 'var' that identifies the specific items (1 = Location; 2 = Price; 3 = Temperature)
gr bar v, over(treat) asyvar over(var) /// Generate the graph with the aesthetic that you wanted 

I tend to think the second method is a bit easier to use and makes it a bit easier to control some of the aesthetics if you're trying to create graphs that will be used in various publications or deliverables.  There  will be a gap between the items in the graph, but no gap between treatment and control groups (although treatment and control will each have their own color).  

HTH,
Billy



On May 31, 2013, at 6:06 AM, Alexis Penot <alexis.penot@ens-lyon.fr> wrote:

> Using by instead of over ?
> 
> Alexis
> 
> Le 31 mai 2013 à 14:58, Andrea Bennett <mac.stata@gmail.com> a écrit :
> 
>> Hi there
>> 
>> I've spent hours on the following issue but seem to be unable to find the correct way to deal with such a graph:
>> 
>> I have questionnaire data where people are ask how much they liked some aspect; e.g. three variables LOCATION, PRICE, TEMPERATURE (all 1-7 scale).
>> 
>> Both the treatment and control group filled in the same questionnaire. I want to see graphically how the treatment and control groups differ in above dimensions.
>> 
>> The following produces the correct data but groups in a "wrong" way. The groups are divided into treated/not-treated:
>> 
>> graph bar location price temperature, over(treated)
>> 
>> However, I would like to have the following:
>> 
>> Location - not treated
>> Location - treated
>> 
>> [gap]
>> 
>> Price - not treated
>> Price - treated
>> 
>> [gap]
>> 
>> Temperature - not treated
>> Temperature - treated
>> 
>> I really don't see who I can produce this result. Options such as -asy- or -asc- seem to go into the right direction but are not producing the result I desire. I would think that this is a rather common feature but it seems utterly difficult to get.
>> 
>> Any suggestions?
>> 
>> Many thanks in advance,
>> 
>> Andrea
>> 
>> 
>> 
>> 
>> *
>> *   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/


*
*   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/