Bar chart comparing rates

[Joanna Davies]

Hello,

Im using the twoway bar command in Stata MP v16.1.

Im trying to create a bar chart comparing rates of hospitalisations for different ethnic groups after adjusting for different variables - so comparing rates by ethnicity in 4 different models, each one controlling for additional characteristics.

The problem is that i want the bars to start at 1 and show the estimated rate above or below 1. The code below gets me bars that start at zero.

Is there a way i can tweak the code below to give me what i need, or do i need to try something else like margins?

I cant post the actual bar chart or data because the data is sensitive - so im hoping the code will give enough information.

My data is long and has 5 variables: ethnicity (9 cats), model (4 cats), irr, lci, uci

There is one additional variable eth_mod which structures the data by model and ethnicity...

gen eth_mod = model if ethnicity==9
replace eth_mod = model+5 if ethnicity==8
replace eth_mod = model+10 if ethnicity==7

[etc - you get the picture re eth_mod so i wont type it all out]

twoway
(bar irr eth_mod if model==1, horizontal) ///
(bar irr eth_mod if model==2, horizontal) ///
(bar irr eth_mod if model==3, horizontal) ///
(bar irr eth_mod if model==4, horizontal) ///
(rcap lci uci eth_mod, horizontal lcolor(black)),
legend(row(1) order(1 "mod1" 2 "mod2" 3 "mod3" 4 "mod4") ///
ylabel(2.5 "eth1) 7.5 "eth2" 12.5 "eth3" 17.5 "eth4" 22.5 "eth5" 27.7 "eth6" 32.5 "eth7" 37.5 "eth8" 42.5 "eth9", noticks angle(horizontal)) ///
ytitle("") xtitle("irr") ///
xline(1, lwidth(thin) lcolor(black) lpattern(solid))

Grateful for any tips.

Many thanks

[Nick Cox]

See the base() option. You want base(1).

I would use scatter || rspike here. with a reference line at 1. And log scale.

[Joanna Davies]

So simple! Thank you!

[Joanna Davies]

Hi Nick,

Returning to the above example, i have another problem. I have added a 5th model and as a result have lost the nice gap between the bars for each of the ethnicity categories. Is there a way to get a small gap between each ethnicity category?

twoway  (bar irr eth_mod if model==1, base(1) horizontal) ///
        (bar irr eth_mod if model==2, base(1) horizontal) ///
        (bar irr eth_mod if model==3, base(1) horizontal) ///
        (bar irr eth_mod if model==4, base(1) horizontal) ///
        (bar irr eth_mod if model==5, base(1) horizontal) ///
        (rcap lci uci eth_mod, horizontal lcolor(black)), ///
        legend(row(1) order(1 "mod1" 2 "mod2" 3 "mod3" 4 "mod4" 5 "mod5")) ///
        ylabel(2.5 "eth1" 7.5 "eth2" 12.5 "eth3" 17.5 "eth4" 22.5 "eth5" 27.5 "eth6" 32.5 "eth7" 37.5 "eth8" 42.5 "eth9", noticks angle(horizontal)) ///
        ytitle("") xtitle("xtitle") xline(1, lwidth(thin) lcolor(black) lpattern(solid)) graphregion(color(white)) title("") ///
        by(group, note("") title("title", size(*0.8)) graphregion(color(white)))

Sorry, again, cant show the actual graph or data...will try to remember to post it on here once its been approved for release.

Thank you,
Joanna

[Joanna Davies]

I should also say that i didnt go down the scatter || rspike route because i quite like the look of the bars. And, the scale is a rate-ratio with a smallish range from about .8 to 1.2 - so i dont think i need to go for a log-scale either. But always grateful for the suggestions.
Best,
Joanna

[Nick Cox]

All you need do is show fake data with the same structure, but for me trying to imagine what that looks like is not going to be fun.

[Andrew Musau]

Would be easier if you added a data example that reproduces your issue.

[Joanna Davies]

Ok, thanks. My data is long and has 5 variables: ethnicity (9 cats), model (5 cats), irr, lci, uci. And one additional variable eth_mod which structures the data by model and ethnicity. Is this enough to go off?

code for gen eth_mod is in #1 the first post

Im not really sure how to create 'fake' data....other than typing out by hand a made up dataset - is there a way to do this quickly? I cant even copy and paste the varnames and categorical variables because its all in a secure settling that wont allow my to copy (sorry!)

[Nick Cox]

You can call up any number of random number functions. This is tough on you, but what otherwise what you're implying is that a reader who can't imagine your problem can only solve it by inventing their own data example or using an existing dataset as sandbox. That might still happen.

[Joanna Davies]

ok, fair point, standby, ive created a fake dataset (it actually didnt take that long!)...just working how to post...

[Joanna Davies]

Ok, here is my attempt at a very basic fake...all the irr lci uci just repeat im afraid - grateful for any advice on how to make it more realistic!

clear
input byte(ethnicity group model) float(irr lci uci eth_mod)
9 1 1 1.5  1   2  1
9 2 1 1.5  1   2  1
9 2 2 1.4 .9 1.9  2
9 1 2 1.4 .9 1.9  2
9 1 3 1.3 .8 1.8  3
9 2 3 1.3 .8 1.8  3
9 1 4 1.2 .7 1.7  4
9 2 4 1.2 .7 1.7  4
9 2 5 1.1 .6 1.6  5
9 1 5 1.1 .6 1.6  5
8 1 1 1.5  1   2  6
8 2 1 1.5  1   2  6
8 1 2 1.4 .9 1.9  7
8 2 2 1.4 .9 1.9  7
8 2 3 1.3 .8 1.8  8
8 1 3 1.3 .8 1.8  8
8 1 4 1.2 .7 1.7  9
8 2 4 1.2 .7 1.7  9
8 1 5 1.1 .6 1.6 10
8 2 5 1.1 .6 1.6 10
7 1 1 1.5  1   2 11
7 2 1 1.5  1   2 11
7 2 2 1.4 .9 1.9 12
7 1 2 1.4 .9 1.9 12
7 1 3 1.3 .8 1.8 13
7 2 3 1.3 .8 1.8 13
7 2 4 1.2 .7 1.7 14
7 1 4 1.2 .7 1.7 14
7 2 5 1.1 .6 1.6 15
7 1 5 1.1 .6 1.6 15
6 1 1 1.5  1   2 16
6 2 1 1.5  1   2 16
6 2 2 1.4 .9 1.9 17
6 1 2 1.4 .9 1.9 17
6 2 3 1.3 .8 1.8 18
6 1 3 1.3 .8 1.8 18
6 2 4 1.2 .7 1.7 19
6 1 4 1.2 .7 1.7 19
6 1 5 1.1 .6 1.6 20
6 2 5 1.1 .6 1.6 20
5 1 1 1.5  1   2 21
5 2 1 1.5  1   2 21
5 1 2 1.4 .9 1.9 22
5 2 2 1.4 .9 1.9 22
5 1 3 1.3 .8 1.8 23
5 2 3 1.3 .8 1.8 23
5 1 4 1.2 .7 1.7 24
5 2 4 1.2 .7 1.7 24
5 1 5 1.1 .6 1.6 25
5 2 5 1.1 .6 1.6 25
4 2 1 1.5  1   2 26
4 1 1 1.5  1   2 26
4 1 2 1.4 .9 1.9 27
4 2 2 1.4 .9 1.9 27
4 2 3 1.3 .8 1.8 28
4 1 3 1.3 .8 1.8 28
4 2 4 1.2 .7 1.7 29
4 1 4 1.2 .7 1.7 29
4 1 5 1.1 .6 1.6 30
4 2 5 1.1 .6 1.6 30
3 2 1 1.5  1   2 31
3 1 1 1.5  1   2 31
3 2 2 1.4 .9 1.9 32
3 1 2 1.4 .9 1.9 32
3 1 3 1.3 .8 1.8 33
3 2 3 1.3 .8 1.8 33
3 2 4 1.2 .7 1.7 34
3 1 4 1.2 .7 1.7 34
3 2 5 1.1 .6 1.6 35
3 1 5 1.1 .6 1.6 35
2 2 1 1.5  1   2 36
2 1 1 1.5  1   2 36
2 2 2 1.4 .9 1.9 37
2 1 2 1.4 .9 1.9 37
2 1 3 1.3 .8 1.8 38
2 2 3 1.3 .8 1.8 38
2 1 4 1.2 .7 1.7 39
2 2 4 1.2 .7 1.7 39
2 1 5 1.1 .6 1.6 40
2 2 5 1.1 .6 1.6 40
1 1 1 1.5  1   2 41
1 2 1 1.5  1   2 41
1 1 2 1.4 .9 1.9 42
1 2 2 1.4 .9 1.9 42
1 2 3 1.3 .8 1.8 43
1 1 3 1.3 .8 1.8 43
1 1 4 1.2 .7 1.7 44
1 2 4 1.2 .7 1.7 44
1 2 5 1.1 .6 1.6 45
1 1 5 1.1 .6 1.6 45
end

[Joanna Davies]

Ah! Ive figured it out...i had forgotten how eth_mod worked...so basically i just needed to change the coding for eth_mod to the following:

gen eth_mod = model if ethnicity==9
replace eth_mod = model+6 if ethnicity==8
replace eth_mod = model+12 if ethnicity==7
replace eth_mod = model+18 if ethnicity==6
replace eth_mod = model+24 if ethnicity==5
replace eth_mod = model+30 if ethnicity==4
replace eth_mod = model+36 if ethnicity==3
replace eth_mod = model+42 if ethnicity==2
replace eth_mod = model+48 if ethnicity==1
sort eth_mod eth_mod ethnicity model

and then reposition the ethnicity labels

A problem shared...many thanks