How to make a graph with confident intervals?

[Matthew Williams]

Dear all,

I would like to make a graph of several concentration indices and their confident intervals against countries. I first use -conindex- command written by O'Donnell and his colleagues to produce values of CIs and confident intervals but I don't how to make a graph like an example below. I would be grateful if someone could help me.

Data

clear
input float(id country y) double inc
5440 2  3  318841.3935546875
5441 2  4    54898.208984375
5442 2  3    54898.208984375
5443 2  1 45175.817321777344
5444 2  0 45175.817321777344
5445 2  0   121893.837890625
5446 2  5   121893.837890625
5447 2  4  71314.00281524658
5448 2  7  71314.00281524658
5449 2  0  38337.68078613281
5450 2  6 11104.256591796875
5451 2  0 11104.256591796875
5452 2  1 26280.073486328125
5453 2  0 26280.073486328125
5454 2  2  28671.24493408203
5455 2  0   787.013916015625
5456 2  3  19524.98480606079
5457 2  0  32819.24786376953
5458 2  5  22196.94139099121
5459 2  0         9848.46875
5460 2  0  34413.07556152344
5461 2  0  34413.07556152344
5462 2  0  23345.85076904297
5463 2  4  23345.85076904297
5464 2  0  25786.55157470703
3050 4  3 13227.870353221893
3051 4  5  6960.977005004883
3052 4  6  6960.977005004883
3053 4  7 3582.0121116638184
3054 4  0  3096.338333129883
3055 4  0  2604.194833278656
3056 4  0 6452.1861572265625
3057 4  2    2447.9052734375
3058 4  0  266.6192626953125
3059 4  3  266.6192626953125
3060 4  0   7081.40771484375
3061 4  1    6782.7939453125
3062 4  0    6782.7939453125
3063 4  0   7038.74853515625
3064 4  6   7038.74853515625
3065 4  0  15041.03581237793
3066 4  7  15041.03581237793
3067 4  0   5154.63916015625
3068 4  3   44727.2001953125
3069 4  0   7891.93017578125
3070 4  2   7891.93017578125
3071 4  3  8306.671447753906
3072 4  4 10622.414672851563
3073 4  1     7380.021484375
3074 4  0     7380.021484375
 800 3  9              16800
 801 3  2              12060
 802 3 10 14492.137939453125
 803 3  2              13200
 804 3  0              13200
 805 3  3              41192
 806 3  0   22124.1123046875
 807 3  5              22450
 808 3  6              22450
 809 3  7  37675.12646484375
 810 3  0              34440
 811 3  0              34440
 812 3  1    104402.56640625
 813 3  0    104402.56640625
 814 3  4              28360
 815 3  0              28360
 816 3  0              10120
 817 3  2  35118.61358642578
 818 3  0  35118.61358642578
 819 3  0      20531.8828125
 820 3  9              12570
 821 3  0              16176
 822 3  0              16176
 823 3  1 38801.344957351685
 824 3  0 34039.176513671875
   1 1  1              23300
   2 1  0              23300
   3 1  8              13200
   4 1  0              23610
   5 1 10              23610
   6 1  8               6586
   7 1  8              30000
   8 1  0              15650
   9 1  1    21901.494140625
  10 1  1 25350.987548828125
  11 1  3              39350
  12 1  5              39350
  13 1  6              26330
  14 1  7              26330
  15 1  0              14800
  16 1  9  21299.12939453125
  17 1  0 18023.751762390137
  18 1  0    19643.787109375
  19 1 10    19643.787109375
  20 1  0                600
  21 1  0                600
  22 1  8       95854.078125
  23 1  2       95854.078125
  24 1  3              23400
  25 1  4              23400
end
label values country countryn
label def countryn 1 "Austria", modify
label def countryn 2 "Switzerland", modify
label def countryn 3 "Belgium", modify
label def countryn 4 "Czech Republic", modify

Code to compute CIs and 95% CI

forval i = 1/4 {
        qui {
            conindex y if country==`i', rankvar(inc) truezero robust    
                gen CI`i' = r(CI)
                gen SE`i' = r(CIse)
                gen df`i' = e(df_r)
                gen lb`i' = CI`i' - invttail(df`i',0.025)*SE`i'
                gen ub`i' = CI`i' + invttail(df`i',0.025)*SE`i'
        }
    }

Thank you

[Nick Cox]

The usual term is confidence interval, not confident interval.

Given your results I would combine the 4 result variables for the index

gen CI = min(CI1, CI2, Ci3, Ci4)

and similarly for ub and lb.

Then you can plot

scatter CI country || rcap ub lb country, xla(1/4, valuelabel noticks)

except it may be more readable to go

scatter country CI || rcap ub lb country, horizontal yla(1/4, ang(h) valuelabel noticks)

[Dung Le]

.

[Matthew Williams]

DearNick Cox,

Thank you for your useful code, they produced nice graph but it seems slightly different from what I am looking for. Since CIs are just values, i.e. CI1= 0.1391489 so I don't know why we need to use its means. Furthermore, a new variable, CI, takes on a negative value while the original values of CI1-CI4 are either positive or negative. As a result, when I made a graph as suggested in #2, the graph does not depict the actual CI values in each country. Is there any other ways you might think of to make a similar graph shown in #1?

This is how a graph looks like using code in #2.


P/s: thanks for your correction regarding the confidence interval term.

[Nick Cox]

The graph looks wrong. That is my fault, because I misread your code in #1. Without revising the code in #1 (I would have advised differently, but no matter) better code would be

gen CI = .  
gen ub = .  
gen lb = .  
forval j = 1/4 {      
      replace CI = CI`j' if country == `j'      
      replace ub = ub`j' if country == `j'      
      replace lb = lb`j' if country == `j'  
}

Then you can plot

 scatter CI country || rcap ub lb country, xla(1/4, valuelabel noticks)

except it may be more readable to go

scatter country CI || rcap ub lb country, horizontal yla(1/4, ang(h) valuelabel noticks)

[Matthew Williams]

DearNick Cox,

Sorry for my late response and thank you so much for your help. The code in #5 works perfectly. Many thanks for that.

Relevant to this thread, do you have any ideas how to make a graph like this? In my data example, I would like to plot combination (stack) of values of CI of age and gender for each country (two countries) in one graph. I would be grateful if you can help me out.



Data

clear
input float(country age gender y wealth)
0 73 1 4    1.395582
0 81 0 5    1.220415
0 62 1 4   1.1294894
0 75 1 6   1.6639395
0 74 1 6     .949276
0 77 1 4   1.6236353
0 61 1 6    .9107408
0 61 1 0   1.2131945
0 72 1 6   1.3741723
0 64 1 1    3.169981
0 75 0 3   1.0510343
0 60 0 0    .8520967
0 68 1 3    .4315759
0 62 1 0   1.5155405
0 65 1 0    .7244153
0 65 1 2   1.1795194
0 71 0 4   2.3465064
0 61 0 0   .12656227
0 77 0 1    .5998135
0 81 0 2    1.048896
0 81 0 3    .6379281
0 66 0 4    .7880797
0 82 0 4    3.020302
0 62 1 3   .20386316
0 85 1 4   .56215984
0 70 1 5   .04049543
0 65 1 6    .9321245
0 67 0 6   .52764887
0 64 1 5    .6974842
0 81 1 0   2.8395345
0 81 1 4    .7161424
0 60 1 4    1.298246
0 81 1 1   2.3377597
0 80 0 6  -.27008873
0 70 1 0   -.2315653
0 60 0 8   -.7111181
0 64 1 8   -.8657792
0 65 1 8   .14117612
0 67 0 2   1.4108043
0 70 1 4   -.2053529
0 74 0 5   -.2418515
0 92 1 3 -.011776933
0 65 1 5   .10739726
0 72 0 5   -.6856656
0 87 1 4    .2986597
0 70 0 2  -.23860455
0 67 0 8   -.7783912
0 85 0 6   .11952727
0 70 0 2  -.58055866
0 68 0 0   .21524943
0 65 1 2  -.26209623
0 77 0 0   -.8062942
0 70 1 4    .3648115
0 62 0 4   .04204537
0 65 1 5   -.4745604
0 71 1 5    .5936369
0 80 1 4   .13920976
0 63 1 2    .7774609
0 85 1 9   1.6758046
0 75 0 3    .5708973
0 72 1 6    .1936778
0 62 1 1   1.0680408
0 71 1 2    .7706921
0 69 1 6   -.1298459
0 61 0 4   .14453334
0 62 0 2   .40860075
0 63 0 4   .09376184
0 66 0 2    .8223537
0 66 0 6     2.49656
0 61 0 4   2.0658643
0 70 1 3   2.6074834
0 65 0 3    2.422002
0 72 1 6    2.410795
0 69 1 1    2.497866
0 60 0 3    .9886292
0 71 0 4   1.4120265
0 82 1 0   .26632148
0 60 1 8   1.2974056
0 73 0 4    .8859935
0 87 1 6   .58373255
0 60 1 4   2.8880665
0 71 0 1   2.0435274
0 82 0 5   2.2385416
0 75 0 3    2.115495
0 71 0 1    2.151605
0 62 0 1    .4529214
0 62 1 3   -.3073213
0 62 1 4    .5955912
0 64 0 1   -.3127069
0 78 0 0   -.4978501
0 62 0 2  .023028206
0 61 0 2    -.422088
0 61 1 4   -.4419667
0 63 1 4   -.3328851
0 67 0 5    .3818175
0 63 0 4   -.2458737
0 73 1 6  -1.3920542
0 69 1 5   -.4999814
0 79 1 6   -.9744884
0 70 0 2   -.4932708
end
label values country country
label def country 0 "A", modify
label values gender gender
label def gender 0 "Male", modify
label def gender 1 "Female", modify

Code to compute CI values for each country

global X "age gender"

foreach x of global X {
    forval i = 0/1 {
        qui {
            conindex `x' if country==`i', rankvar(wealth) truezero robust
                sca CI_`x'_`i' = r(CI)*1000
        }
                sca di CI_`x'_`i'
    }
}

Results

CI_age_0 =  1.0900621
  CI_age_1 = -6.8635584
CI_gender_0 = -6.0384654
CI_gender_1 = -27.653923

Thank you.

[Matthew Williams]

Dear Nick Cox,

I hope you are doing great!

I am writing this because I am not sure if you have missed information in #6.

Thank you.

[Nick Cox]

Either I missed the ping or I didn't understand the question in #5, which refers to a community-contributed program conindex from the Stata Journal. I suggest that you use statsby to put results in new variables, but even then I don't see any connection between the graph in your image and the results you might get.

I'd guess very many readers here would interpret CI as confidence interval. It seems that you mean "concentration index".

[Matthew Williams]

Dear Nick Cox,

Thank you for your response as well as your suggestions. You are right, CI in #6 refers to concentration index estimated from -conindex- command which I mentioned in #1.
Let me explain what in #6 meant. I was calculating the values of concentration index (CI) for age and gender in each country and then I wanted to plot those values for each country in one graph as illustrated by an image example in #6. For example, the total CI of country 1 would equal age1 + gender 1; country2 = age2 + gender2 and so on. I hope that this is clear to you.
Data

clear
input float(id y wealth) int age byte gender float country
5795  4 4 67 0 2
5796  5 4 67 1 2
5797  6 2 63 1 2
5798  0 5 61 1 2
5799  7 5 73 0 2
5800  8 3 56 1 2
5801  9 2 64 0 2
5803 10 1 52 1 2
5804 11 1 61 1 2
5805  0 1 56 0 2
5806  0 5 64 0 2
5807  0 5 62 0 2
5808 12 5 56 1 2
5809 13 3 63 1 2
5810  3 2 73 0 2
5811  0 1 64 1 2
5812  1 1 58 0 2
5813  0 4 64 0 2
5814  1 2 51 1 2
5815  1 3 55 1 2
5816  2 3 59 1 2
5817  3 2 64 1 2
5818  0 3 73 0 2
5819  0 3 75 1 2
5820  0 2 57 0 2
3696 12 4 66 1 4
3697  2 4 73 0 4
3698  4 5 57 1 4
3699  5 3 70 0 4
3700  6 3 56 1 4
3701  7 5 60 1 4
3702  8 4 58 0 4
3703  9 4 51 1 4
3704  0 5 61 1 4
3705  0 4 64 0 4
3706 10 4 65 1 4
3707  2 2 60 0 4
3708  1 2 59 1 4
3709 11 2 64 1 4
3710  0 1 67 1 4
3711  2 4 53 1 4
3712 13 4 54 0 4
3713  3 3 57 1 4
3714  0 3 64 0 4
3715  3 5 60 1 4
3716  1 5 61 0 4
3717  0 5 50 1 4
3718  3 5 53 1 4
3719  1 5 56 0 4
3720  2 5 71 1 4
1161  4 3 52 1 3
1162  5 2 54 1 3
1163  6 1 51 0 3
1164  7 4 56 0 3
1165  8 4 53 1 3
1166  9 1 63 1 3
1167 10 4 60 0 3
1168 11 4 50 1 3
1169  0 2 52 1 3
1170  4 1 54 0 3
1171  0 1 54 1 3
1172  0 3 52 1 3
1173 12 1 61 1 3
1174  0 1 57 0 3
1175  0 4 73 1 3
1176  0 4 75 0 3
1177 13 3 52 0 3
1178  3 1 51 1 3
1179  0 1 53 0 3
1180  0 1 50 1 3
1181 13 2 68 0 3
1182  2 2 61 1 3
1183  0 5 70 0 3
1184  0 5 65 1 3
1185  0 3 51 0 3
   1  2 2 54 0 1
   2  1 2 51 1 1
   3  4 5 58 0 1
   4  5 2 73 1 1
   5  6 1 56 0 1
   6  7 1 53 1 1
   7  8 3 71 1 1
   8  9 5 74 1 1
   9 10 5 69 0 1
  10 11 4 67 0 1
  11 12 1 70 0 1
  12  1 1 67 1 1
  13  0 3 64 1 1
  14 13 3 69 0 1
  15  3 3 65 1 1
  16  0 4 65 0 1
  17  2 5 61 0 1
  18  0 3 66 0 1
  19  5 3 62 1 1
  20  0 2 63 0 1
  22  0 5 54 1 1
  23  0 3 66 0 1
  24  0 3 60 1 1
  25  0 1 68 0 1
  26  0 4 60 1 1
end
label values country countryn
label def countryn 1 "Austria", modify
label def countryn 2 "Switzerland", modify
label def countryn 3 "Belgium", modify
label def countryn 4 "Czech Republic", modify

Code

global X "age gender"
gen age1=.
gen age2=.
gen age3=.
gen age4=.

gen gender1=.
gen gender2=.
gen gender3=.
gen gender4=.

foreach x of global X {
    forval i = 1/4 {
        qui {
            conindex `x' if country==`i', rankvar(wealth) truezero robust
                sca CI_`x'`i' = r(CI)*1000
        }
                sca di CI_`x'_`i'
                replace `x'`i' = CI_`x'`i' if country==`i'
    }
}

Thank you

[Nick Cox]

Too late now, but this from #6 onwards should have been a different thread

My guess is that graph hbar with stacking is what you need.

[Amelie Nickel]

Confidence Interval with Graph Dot

Dear all,
I am dealing with a similar issue. I want to add confidence intervals to a graph dot showing the mean value of anti-immigrant attitudes for each country.

Based on cross-sectional data including 28 countries, i generated the following example dataset consisting of the mean value of anti-immigrant attitudes for each country.
----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. To install: ssc install dataex
clear
input float antmi str2 cntry
 5.179465 "AT"
4.4687862 "BE"
 5.980606 "BG"
 3.984998 "CH"
 5.756306 "CY"
 6.139078 "CZ"
 4.171445 "DE"
 4.239356 "DK"
 5.185469 "EE"
3.9809716 "ES"
 3.886236 "FI"
 4.790751 "FR"
4.1734457 "GB"
 4.947642 "HR"
 6.218821 "HU"
 3.810525 "IE"
2.7447665 "IS"
 5.511042 "IT"
 4.952845 "LT"
 4.521533 "LV"
4.1681414 "NL"
 3.972714 "NO"
 4.316811 "PL"
 3.811007 "PT"
 5.703016 "RS"
 3.627834 "SE"
 5.487343 "SI"
 6.264863 "SK"
end

------------------ copy up to and including the previous line ------------------

The graph dot attached shows the mean value of anti-immigrant attitudes (dots) over countries (categorial axis presented vertically) and is based on the following command

. graph dot antmi, over (cntry, sort(1)descending) ///
>         yline (5.09, lcolor(gray) lpattern(dash)) ///
>         ytitle ("Anti-Immigrant Attitudes over country")

I want to add confidence intervals to the graph dot for each country mean of anti-immigrant attitudes similar to Figure 1 in #1.

I would be very grateful for any help.

Best regards
Amelie