Creating composite index/Variable

[Sonwabile Mbuma]

Hi. MAY YOU PLEASE ASSIST.

I am trying to do two things:

1. To create a composite index/variable from 4 individual variables of a dataset (i.e. create an outcome variable from 4 individual variables). This will help with my multinomial regression model.

2. To have a STATA code that will measure levels based on the number of correct responses (i.e. None=0, Low=1-2, Average=3, High=4-5)...such that, I can be able to say x% of respondents are at a low level because they only answered 1 or 2 questions correctly.

I will really appreciate...thanks in advance.

[Andrew Musau]

Can you present a data example, e.g., copying and pasting the output of

dataex

?

[Sonwabile Mbuma]

. dataex
input statement exceeds linesize limit. Try specifying fewer variables
r(1000);

When I specified the variable I need to use to create an a composite variable:

label values q3_1b
label def q3_1b 1 "Agree", modify
label def q3_1b 2 "Disagree", modify
label def q3_1b 3 "Do not know", modify
label values q3_1c
label def q3_1c 1 "Agree", modify
label def q3_1c 2 "Disagree", modify
label def q3_1c 3 "Do not know", modify
label values q3_2f
label def q3_2f 1 "Agree", modify
label def q3_2f 2 "Disagree", modify
label def q3_2f 3 "Do not know", modify
label values q3_3e
label def q3_3e 1 "Yes", modify
label def q3_3e 2 "No", modify
label def q3_3e 3 "Do not know", modify
label values q3_4
label def q3_4 1 "Yes", modify
label def q3_4 2 "No", modify
label def q3_4 3 "Do not know", modify

[Sonwabile Mbuma]

I am very new on stata.

[Andrew Musau]

Try

qui ds
dataex `=word("`r(varlist)'", 1)' - `=word("`r(varlist)'", 5)'

If that fails, change the highlighted "5" to "4", then run the code again. If that fails, change it to "3", and so on.

[Sonwabile Mbuma]

. qui ds

.
. dataex `=word("`r(varlist)'", 1)' - `=word("`r(varlist)'", 5)'

----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. To install: ssc install dataex
clear
input double idnum str7 barcode long sal byte(persno province)
      130000 "SV02520" 1993605 . 1
     5008000 "SV34432" 5920215 8 5
     7984253 "SV36954" 7984253 . 7
     7984253 "SV37000" 7984253 . 7
     7984999 "SV37943" 7984999 . 7
     7985338 "SV36983" 7985338 . 7
    24004000 "SV06387" 5992646 4 5
   101003000 "SV08474" 7984253 3 7
   125006000 "SV04767" 1991704 6 1
   182001000 "SV08501" 7984999 1 7
   251002000 "SV08514" 7984999 2 7
   251004000 "SV08516" 7984999 4 7
   278001000 "SV08401" 7984735 1 7
   290070000 "SV31926" 7984253 . 7
   311070000 "SV36921" 7600909 . 7
   355210000 "SV36942" 7984253 . 7
 19936430000 "SV34347" 1993643 . 1
 36001470000 "SV04989" 3600147 . 3
 4.72004e+10 "SV30649" 4720040 . 4
 47400430000 "SV90336" 4740043 . 4
 56900610000 "SV22414" 5690061 . 5
 56901230000 "SV35358" 5690123 . 5
 58302140000 "SV52999" 5830214 . 5
 59947490000 "SV03782" 5994749 . 5
 86100930000 "SV51059" 8610093 . 8
 86101170000 "SV22008" 8610117 . 8
 86300150000 "SV51052" 8630015 . 8
 86400990000 "SV65770" 8640099 . 8
 86401610000 "SV51058" 8640161 . 8
 86602440000 "SV11891" 8660244 . 8
 86604440000 "SV11934" 8660444 . 8
 86804080000 "SV13872" 8680408 . 8
160004000101 "SV11221" 1600040 1 1
160004000102 ""        1600040 2 1
160004000103 "SV19655" 1600040 3 1
160004000104 "SV14452" 1600040 4 1
160004000105 "SV10511" 1600040 5 1
160004000301 ""        1600040 1 1
160004000503 ""        1600040 3 1
160004000603 ""        1600040 3 1
160004000803 ""        1600040 3 1
160004000903 ""        1600040 3 5
160004001101 "SV12234" 1600040 1 1
160004001101 "SV65327" 1600040 1 1
160004001102 "SV65313" 1600040 2 1
160004001103 ""        1600040 3 1
160004001501 ""        1600040 1 1
160004001601 "SV10465" 1600040 1 1
160004001602 "SV26120" 1600040 2 1
160004002101 "SV10481" 1600040 1 1
160004002301 ""        1600040 1 1
160004002701 "SV65239" 1600040 1 1
160004002702 ""        1600040 2 1
160004002703 "SV65296" 1600040 3 1
160004003201 "SV65301" 1600040 1 1
160004003202 "SV65306" 1600040 2 1
160004003203 "SV31053" 1600040 3 1
160004003204 "SV32985" 1600040 4 1
160004003405 ""        1600040 5 1
160004003405 ""        1600040 5 1
160004003701 "SV65258" 1600040 1 1
160004003702 "SV65333" 1600040 2 1
160004003703 "SV65340" 1600040 3 1
160004003704 "SV52297" 1600040 4 1
160004004201 "SV04715" 1600040 1 1
160004004202 "SV65315" 1600040 2 1
160004004203 "SV65278" 1600040 3 1
160004004204 "SV65336" 1600040 4 1
160004004205 "SV52352" 1600040 5 1
160004004206 "SV67028" 1600040 6 1
160004004701 "SV65202" 1600040 1 1
160004004702 "SV65324" 1600040 2 2
160004004702 "SV65324" 1600040 2 1
160004004703 "SV52200" 1600040 3 1
160004004703 ""        1600040 3 2
160004004704 ""        1600040 4 1
160004004705 ""        1600040 5 1
160004005301 "SV10455" 1600040 1 1
160004005801 "SV08272" 1600040 1 1
160004005802 "SV65240" 1600040 2 1
160004005803 ""        1600040 3 1
160004006301 "SV65325" 1600040 1 1
160004006801 "SV65330" 1600040 1 1
160004006802 "SV65339" 1600040 2 1
160004006803 "SV52301" 1600040 3 1
160004008002 "SV13183" 1600040 2 1
160004012301 ""        1600040 1 1
160004013403 ""        1600040 3 1
160004014502 ""        1600040 2 1
160004016503 "SV19991" 1600040 3 2
160004023404 ""        1600040 4 1
160004032103 ""        1600040 3 1
162001800501 "SV19417" 1620018 1 1
162001800502 "SV22858" 1620018 2 1
162001800503 "SV11236" 1620018 3 1
162001800504 "SV10582" 1620018 4 1
162001800505 ""        1620018 5 1
162001800701 "SV65671" 1620018 1 1
162001800702 ""        1620018 2 1
162001801001 "SV65633" 1620018 1 1
end
label values province province
label def province 1 "Western Cape", modify
label def province 2 "Eastern Cape", modify
label def province 3 "Northern Cape", modify
label def province 4 "Free State", modify
label def province 5 "KwaZulu-Natal", modify
label def province 7 "Gauteng", modify
label def province 8 "Mpumalanga", modify

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

Listed 100 out of 66615 observations
Use the count() option to list more

[Sonwabile Mbuma]

On my second question: No Knowledge = Those who answered all 5 questions incorrectly, combined with those who said they did not know all question = 3235 + 246......Similarly, Those who answered all questions correctly are 3235 participants. How do I count those who answered 1, 2, 3, or 4 questions correctly?

. count if q3_1b ==1 & q3_1c ==1 & q3_2f ==1 & q3_3e ==1 & q3_4 ==1
3,235

. count if q3_1b ==2 & q3_1c ==2 & q3_2f ==2 & q3_3e ==2 & q3_4 ==2
246

. count if q3_1b ==3 & q3_1c ==3 & q3_2f ==3 & q3_3e ==3 & q3_4 ==3
1,232

Truth values: 1=True, 2=false; 3=dont know

[Nick Cox]

Of your questions in #1

1. To create a composite index/variable from 4 individual variables of a dataset (i.e. create an outcome variable from 4 individual variables).

Sorry, but I have no idea what that means precisely. A composite variable could be (a) a mean (b) a median (c) some other summary (d) a score from factor analysis, principal component analysis or correspondence analysis (e) something else.

2. To have a Stata code that will measure levels based on the number of correct responses (i.e. None=0, Low=1-2, Average=3, High=4-5)...such that, I can be able to say x% of respondents are at a low level because they only answered 1 or 2 questions correctly.

Your post #3 indicates that you have variables such as

q3_1b q3_1c q3_2f q3_3e q3_4 some of which have values 1 "Agree" 2 "Disagree" 3 "Do not know"

others of which have values 1 "Yes" 2 "No" 3 "Do not know"

Your post #7 complicates matters further by mentioning 1=True, 2=false; 3=dont know

So, what defines correct? It seems that you mean answering 1.

Here is some technique that may help. If not, I suspect you need to be much clearer on what you have and what you want.

* Example generated by -dataex-. For more info, type help dataex
clear
input float(id q1 q2 q3)
1 1 2 1
2 1 2 1
3 1 2 2
4 1 3 3
5 2 3 3
end

egen count1 = anycount(q*), values(1)

egen count2 = anycount(q*), values(2)

egen count3 = anycount(q*), values(3)

list

     +----------------------------------------------+
     | id   q1   q2   q3   count1   count2   count3 |
     |----------------------------------------------|
  1. |  1    1    2    1        2        1        0 |
  2. |  2    1    2    1        2        1        0 |
  3. |  3    1    2    2        1        2        0 |
  4. |  4    1    3    3        1        0        2 |
  5. |  5    2    3    3        0        1        2 |
     +----------------------------------------------+

[Sonwabile Mbuma]

Thank you very much Nick!! The code works so well. It is exactly what I will need for the second part of my problem; which is determining how many people answered (how many questions) correctly?


Just to clarify the first part of my problem: I need to create one variable from q3_1b q3_1c q3_2f q3_3e q3_4 (which will be my dependent variable). This will be a way of compiling one score from the questions or statements that represent knowledge (i.e. q3_1b q3_1c q3_2f q3_3e & q3_4). The variable will be then used in a multinomial regression analysis (looking at factors affecting knowledge).

I hope this clarifies what I need to do. I am not quite sure whether this is doable in ways that do not reduce the level of contribution of each question towards the index (the variable that i need to create).

[Sonwabile Mbuma]

Thank you very much Nick!! The code works so well. It is exactly what I will need for the second part of my problem; which is determining how many people answered (how many questions) correctly?


Just to clarify the first part of my problem: I need to create one variable from q3_1b q3_1c q3_2f q3_3e q3_4 (which will be my dependent variable). This will be a way of compiling one score from the questions or statements that represent knowledge (i.e. q3_1b q3_1c q3_2f q3_3e & q3_4). The variable will be then used in a multinomial regression analysis (looking at factors affecting knowledge).

I hope this clarifies what I need to do. I am not quite sure whether this is doable in ways that do not reduce the level of contribution of each question towards the index (the variable that i need to create).

[Nick Cox]

Good to hear of progress on your second question. Otherwise your explanation does not take me beyond your initial statement that you want to calculate a composite score. As already detailed in #8, there are many ways to do that.

[Sonwabile Mbuma]

I am not absolutely sure which one is best suited. But I strongly believe that a composite that gives the mean of all five variables is what I need.

[Nick Cox]

Use egen then. It has a rowmean() function.

[Sonwabile Mbuma]

Thank you.