st: AW: RE: AW: RE: AW: AW: RE: -ciplot- -eclplot- combine with box-options?

["Kaulisch, Marc" <kaulisch@forschungsinfo.de>]

Nick, you are right. I was not not clear enough ( I hope the e-mail-programs show the tabs clear enough):

I would like to see the confidence-intervals (boxes) for each of the categories and combination of categories in vertical order. I think this more readable as if the plots for each first category (Cat1) are separately plotted. Especially, if I would like to spot not only differences between one category (Cat2) but also to see whether items in category 1 are different. This is less a problem with "sex" as a by-variable but if I have eight items in a by-var it is very hard to tell differnces...

I hope the examples in my earlier post show this...

My example 1 shows the categories like this:
	cat2_1      box
Cat1_1 
	cat2_2	box
 
	cat2_1	box
Cat1_2
	cat2_2	box
...

My example 2 shows the categories like:
CompCat_1_1		box
CompCat_1_2		box
CompCat_2_1		box
CompCat_2_2		box

With By() I get
		Cat1_1		Cat1_2
Cat2_1	box			box
Cat2_2	box			box


Marc


-----Ursprüngliche Nachricht-----
Von: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Nick Cox
Gesendet: Montag, 10. Mai 2010 12:31
An: statalist@hsphsun2.harvard.edu
Betreff: st: RE: AW: RE: AW: AW: RE: -ciplot- -eclplot- combine with box-options?

My comment was a lament, not a reproach. I'll look at the help for
-stripplot- to make it clearer that -by()- is an easy alternative to -over()-. 

In return, I am not clear what you mean by better interpretable graphs than those generated with -by()-. 

Nick
n.j.cox@durham.ac.uk 

Kaulisch, Marc

Dear Nick,

Thanks for the follow up clarifing opportunities with -stripplot-. I am not as much into Stata that I can adhoc find solutions for two nice-to-haves I would like. 1) Avoid a composite variable for including two categorical variables (comparing sex by agegroup). 2) Better interpretable graphs than those generated with by().

I might miss features of -stripplot- or Stata again.

. sysuse bpwide, clear

1) Example with boxplot
. graph hbox bp_before bp_after , over(sex, label(labs(vsmall))) over(agegrp, label(labs(vsmall)) sort(1) descending)

2) Example with Confidence-intervals:

. tempfile tf1 tf2
. statsby mean1=r(mean) ub1=r(ub) lb1=r(lb) N1=r(N), by(agegrp)
saving(`tf1'): ci bp_before
. statsby mean2=r(mean) ub2=r(ub) lb2=r(lb) N2=r(N), by(agegrp)
saving(`tf2'): ci bp_after
. dsconcat `tf1' `tf2'
. generate axis = 6 if agegrp == 3 & sex == 0 . replace axis = 5 if agegrp == 3 & sex == 1 . replace axis = 4 if agegrp == 2 & sex == 0 . replace axis = 3 if agegrp == 2 & sex == 1 . replace axis = 2 if agegrp == 1 & sex == 0 . replace axis = 1 if agegrp == 1 & sex == 1 . la de laxis 6 "60+ Male" 5 "Female" 4 "46-59 Male" 3 "Female" 2 "30-45 Male" 1 "Female"
. la val axis laxis
. twoway scatter  axis mean1 || rcap ub1 lb1 axis, hori || scatter  axis
mean2 || rcap ub2 lb2 axis, hori , ///
			ylabel(1(1)6, labs(vsmall) nogrid val
angle(hori))			///	
			ytitle("")
///
			legend(label(1 "Mean bp_before") label(2 "CI
bp_before") ///
			label(3 "Mean bp_after") label(4 "CI bp_after")
size(vsmall) rows(2) span)
			

3) Stripplot (!?) with Confidence intervals:
* Generate axis as in solution (2)
. stripplot bp_before bp_after , by(axis, compact) vertical bar
boffset(0) ms(none) xsc(r(0.8 2.2))

Nick Cox

It's a pity that Marc missed a (simpler) solution with -stripplot- that requires no dataset manipulation at all. 

sysuse bpwide 

stripplot bp_before bp_after , by(sex, compact) stack vertical bar
height(0.2)

gets you there directly. 

If you really want only the c.i. bars, so that the data points are suppressed, that's perverse, but you can do it too: 

stripplot bp_before bp_after , by(sex, compact) vertical bar boffset(0)
ms(none) xsc(r(0.8 2.2))

In short, -stripplot- for two or more variables and two or more groups requires -by()- not -over()-. As program writer, I saw no point trying to re-create a complicated set-up when -by()- does it for you for free. 

Nick
n.j.cox@durham.ac.uk 

Kaulisch, Marc

Okay, it seems that -statsby- does the trick together with -dsconcat- (ssc).

Preliminary without labels etc:

sysuse bpwide
tempfile tf1 tf2
statsby mean1=r(mean) ub1=r(ub) lb1=r(lb) N1=r(N), by(agegrp)
saving(`tf1') subsets total: ci bp_before statsby mean2=r(mean)
ub2=r(ub) lb2=r(lb) N2=r(N), by(agegrp)
saving(`tf2') subsets total: ci bp_after dsconcat `tf1' `tf2'
twoway rcap ub1 lb1 agegrp || scatter mean1 agegrp || rcap ub2 lb2 agegrp || scatter mean2 agegrp 

Kaulisch, Marc

Nick,

Thanks a lot for this extensive answer. I learn a lot. I haven't had time yet to go deeper in the opportunities to work with statsby (will do this later). But experimenting with -stripplot- I come very early to an end...

I would like to create a graph such as
. sysuse bpwide
. graph hbox bp_before bp_after, over(sex) over(agegrp), sort(1)
descending)

But I do not want a boxplot but a ci-bar as in stripplot...

-stripplot- does not allow two vars with over.

. ciplot bp_before bp_after, by(agegrp)
looks closer to my interest...

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