Histogram in stata

[amanda rasmussen]

Hi.

I have a question:
How do I make a histogram with a variable on the x-axis and a variable on the y-axis?
When I type
histogram mean_ci gestationsdage
stata says "too many variables specified".

When I make a scatterplot instead, there are no problem.
I hope someone could help!

Regards

[Nick Cox]

What would that look like?

A perspective view of a two-dimensional array of bars?

A heat map shaded according to frequencies or on some other scale?

Histograms along axes and something else in the main plot region?

It's probably best to give us an example of your data, and then we can suggest graphics.

Please do read and act on https://www.statalist.org/forums/help#stata

[amanda rasmussen]

I would like to illustrate the distribution of some cardiac measurements in different gestational ages. On the x-axis I want gestational age, and on the y-axis I want the cardiac measurements. Both variables are continious variables.
Thanks.

[Nick Cox]

Sorry, but "illustrate the distribution" is too vague for me to offer precise advice. Or rather I could add lots of precise advice but most of it would miss your target.

Do you mean that gestational age could be 12.345678 weeks?

As before, please give an example of data. If your data are confidential, fake data with the same form are fine. Just add some random numbers to your real data.

[Marcos Almeida]

Just a side note, after Nick's advice.

I would like to illustrate the distribution of some cardiac measurements in different gestational ages. On the x-axis I want gestational age, and on the y-axis I want the cardiac measurements. Both variables are continious variables.

Please check the examples in the Stata Manual, after typing - help histogram - for that matter. That said, if you have two continuous variables and you wish to "illustrate" the relationship between both, I fear histograms are not that best approach. What is more, dutifully enough, the command - histogram - wouldn't allow for more than one variable at a time, and that is the reason Stata gives the error message "too many variables specified".

[amanda rasmussen]

* Example generated by -dataex-. To install: ssc install dataex
clear
input int gestationsdage float(mean_svr mean_co)
280 1009.38431 7.0147291
280 1113.97426 6.4385723
280 1290.25242 5.9834751
271 923.85481 6.9160694
271 998.0857 7.2956475
285 891.11355 7.0701717
285 1159.87458 5.9019181
285 1285.2488 5.5834116
274 619.90545 9.316148
274 894.50551 9.1746449
274 872.97065 9.2963243
284 930.08871 7.542786
287 1112.14076 5.4396524
287 1312.0278 5.4251245
287 1197.26883 5.6842894
292 1061.6838 6.6624285
292 1158.70486 6.0765824
292 1274.70582 5.7765065
292 1407.1181 5.342857
292 1127.68824 6.1843597
292 1042.2873 7.1423317
292 1165.33867 6.364958
284 1011.60871 7.4328362
284 1052.18565 6.7220922
284 1148.29635 6.6285944
284 1261.326767 6.3857242
278 1026.86268 6.4684452
278 1097.15367 6.284096
278 1101.57865 6.498024
273 1124.61856 6.9270183
273 1028.93602 7.5955082
273 1182.246656 6.5343572
287 1006.91069 7.3063867
287 993.23862 7.378155
287 973.72361 7.57230334
277 908.3795 6.7880173
277 1022.31063 6.5468634
277 1031.52064 6.622535
269 940.43641 6.4184656
269 1080.17698 6.6051416
269 1152.09645 6.0557436
286 961.32604 7.0660438
286 991.79545 7.3831628
286 1121.92235 6.8863119
286 1109.81932 7.4125111
end
[/CODE]

[amanda rasmussen]

Thanks Marcos. Unfortunately, I have been asked to illustrate it with a histogram...

[Nick Cox]

OK, so your gestational age is in days, presumably, but what you're showing here are already means of some kind. Sorry, yet more questions, but you aren't giving much away:

1. What's the total size of the dataset?

2. Do you want distributions for each distinct age (#days), or are you willing to bin?

3. What is the range of gestational age in your complete dataset?

4. What are your means over? Replicate measurements for each individual baby? Or something else?

I am not a clinician or a medical statistician, but I assuming that this is all about distributions conditional on gestational age, so that is the focus rather than the joint distribution.

[amanda rasmussen]

Yes, gestational age are in days.
1: 346 persons are included in the dataset.
2: willing to bin
3: 238 days to 316 days
4: The mean_svr and mean_co are the mean over replicate heart function measurements of the mother.

[Nick Cox]

OK. Thanks. So you have about 8 times more data than you show us. I am inclined to regard this as a matter of smoothly changing distribution rather than binning.

Here is an adaptation of an example in the help of rangestat (SSC). The idea is to show moving quantiles and the data too.

It is adaptable in that you can tune which quantiles and which window width you choose. The example shows 7 day windows.

It is not smart about weighting within windows (meaning, there is none).

In broad terms, the scatter here is necessarily quite large given other uncontrolled variables. No one should believe in the dip from 273 to 274 (and that age is an estimate in most cases, presumably??).

With more data, the results should be less irregular, but you may still need to smooth more.

* Example generated by -dataex-. To install: ssc install dataex
clear
input int gestationsdage float(mean_svr mean_co)
280 1009.38431 7.0147291
280 1113.97426 6.4385723
280 1290.25242 5.9834751
271 923.85481 6.9160694
271 998.0857 7.2956475
285 891.11355 7.0701717
285 1159.87458 5.9019181
285 1285.2488 5.5834116
274 619.90545 9.316148
274 894.50551 9.1746449
274 872.97065 9.2963243
284 930.08871 7.542786
287 1112.14076 5.4396524
287 1312.0278 5.4251245
287 1197.26883 5.6842894
292 1061.6838 6.6624285
292 1158.70486 6.0765824
292 1274.70582 5.7765065
292 1407.1181 5.342857
292 1127.68824 6.1843597
292 1042.2873 7.1423317
292 1165.33867 6.364958
284 1011.60871 7.4328362
284 1052.18565 6.7220922
284 1148.29635 6.6285944
284 1261.326767 6.3857242
278 1026.86268 6.4684452
278 1097.15367 6.284096
278 1101.57865 6.498024
273 1124.61856 6.9270183
273 1028.93602 7.5955082
273 1182.246656 6.5343572
287 1006.91069 7.3063867
287 993.23862 7.378155
287 973.72361 7.57230334
277 908.3795 6.7880173
277 1022.31063 6.5468634
277 1031.52064 6.622535
269 940.43641 6.4184656
269 1080.17698 6.6051416
269 1152.09645 6.0557436
286 961.32604 7.0660438
286 991.79545 7.3831628
286 1121.92235 6.8863119
286 1109.81932 7.4125111
end

* ssc inst moremata needed for -mm_quantile()- 
mata:  
mata clear
real rowvector myquantile(real colvector X) {
     if (rows(X) < 3) return(.) 
     return(mm_quantile(X, 1, (0.75, 0.5, 0.25)))
}
end 

rangestat (myquantile) mean_svr, interval(gest -3 3) 

label var myquantile3 "p25"
label var myquantile2 "p50"
label var myquantile1 "p75"

set scheme s1color 
scatter mean_svr gest, ms(oh) mc(gs8) || ///
line myquantile? gest, sort legend(order(2 3 4) col(1) pos(5) ring(0)) ///
            ytitle("mean_svr") yla(, ang(h))

[amanda rasmussen]

Thank you very much Nick for your help!

[Nick Cox]

Here's another way to approach it with stripplot (SSC). I choose bin width 4 days (lower limits shown) for no reason at all.

. gen bin = 4 * floor(gest/4)

. tab bin

        bin |      Freq.     Percent        Cum.
------------+-----------------------------------
        268 |          5       11.11       11.11
        272 |          6       13.33       24.44
        276 |          6       13.33       37.78
        280 |          3        6.67       44.44
        284 |         18       40.00       84.44
        292 |          7       15.56      100.00
------------+-----------------------------------
      Total |         45      100.00

. stripplot mean_svr , over(bin) stack vertical height(1) width(20) ms(Sh)

[amanda rasmussen]

Thanks again. I like this graph!

[Nick Cox]

dotplot as an official command should be mentioned.

[Marcos Almeida]

I have a question:
How do I make a histogram with a variable on the x-axis and a variable on the y-axis?

[...] I have been asked to illustrate it with a histogram...

I agree the graphics provided in #10 ad #12 are rather insightful solutions. That said, at least for the sake of debate, I wonder whether a "histogram" with two continuous variables, one in each axis, as demanded, wouldn't still be taken as an impossible mission, conceptually speaking, given the conflict with the "solitude" of a continuous variable, necessary to provide histograms.