Categorising and generating new variables based on Likert Scale variables

[Belcaryn Catanza]

Hello,
I was hoping someone could help me please.

I am doing my thesis on Inter-Parental conflict and children's mental health. I am going to investigate the differential effects that Verbal IPC, Physical IPC and Co-occurring IPC has on children's mental health as measured by the Argumentative Relationship Scale which has 5 items
and 5 responses per item/

5 item variables Value Labels
are15a1 = 1 -- Never, 2 --Rarely, 3 -- Sometimes, 4 --Often, 5 --Always (Verbal IPC)
are15a2 = 1 -- Never, 2 --Rarely, 3 -- Sometimes, 4 --Often, 5 --Always (Verbal IPC)
are15a3 = 1 -- Never, 2 --Rarely, 3 -- Sometimes, 4 --Often, 5 --Always (Verbal IPC)
are15a4 = 1 -- Never, 2 --Rarely, 3 -- Sometimes, 4 --Often, 5 --Always (Verbal IPC)
are15a5 = 1 -- Never, 2 --Rarely, 3 -- Sometimes, 4 --Often, 5 --Always (Physical IPC)

e.g.,
are15a1 = How often would you and your partner have disagreements regarding child rearing issues?

Please note: co-occurring IPC is represented by a response of Rarely - Always on ANY of the first 4 items and Rarely-Always on the last item


Firstly, I need to establish Verbal IPC, Physical IPC and Co-Occurring IPC variables. But having trouble doing so.

I type in the below code and it gives me an invalid 'are15a2' error so i remove are15a2 just to see what it does, and it gives me invalid 'are15a2' error

Could you please help me?

gen Verbal_IPC = 1 if are15a1 are15a2 are15a3 are15a4 !="1 -- Never" & if are15a5 =="1 -- Never"
gen Physical_IPC = 2 if are15a1!="1--Never"
gen Co_IPC = 3 if are15a1 are15a2 are15a3 are15a4 !="1--Never" & if are15a5 !=="1--Never"


Thank you, really appreciate your help!

[Harrison Ochieng]

So, just to get it right...the new variable Verbal_IPC will include are15a1, are15a2, are15a3, are15a4?

[Nick Cox]

If 5 variables each have 5 possible categories, then the number of possible cross-combinations is evidently 5^5 = 3125. .

What you should know is that

are15a1 are15a2 are15a3 are15a4 !="1--Never"

is not a way to say in Stata (or in any language???)

are15a1 != "1--Never" & are15a2 != "1--Never" & are15a3 != "1--Never" & are15a4 != "1--Never"

even if those 4 are string variables. If "1--Never" is as said, a value label, then this would be one way to work. :

are15a1 != 1  & are15a2 != 1 & are15a3 != 1 & are15a4 != 1

assuming that "1--Never" is the value label corresponding to value 1.

[Belcaryn Catanza]

Hello Harrison, Yes. The Verbal IPC variable includes a response of Rarely - Always on ANY of the first 4 items (are15a1, are15a2, are15a3, are15a4)

[Belcaryn Catanza]

Thank you Nick. Yes, "1--Never" is the value label corresponding to value 1. I have copied the value labels below for your reference

1 -- Never
2 -- Rarely
3 -- Sometimes
4 -- Often
5 -- Always


I will try the code you suggested and will let you know how i get on. Thank you

[Nick Cox]

There are various tricks here.

1. Use egen to concatenate. Then

egen area15a = concat(area15a?)

gives you string variables with values from "11111" to "55555"

so that your condition that no variables have value 1 is equivalent to

strpos(area15a, "1") == 0

However, watch out for missing values.

2. Other egen functions can be useful, including rowmin() rowmax() rowtotal() anycount() anymatch() anyvalue().

[Belcaryn Catanza]

Hey Nick,

The below code worked. I just need to put the OR command in, as a reponse of Always - Never on ANY of the first 4 variables (are15a1, are15a2, are15a3, are15a4) represent Verbal IPC..

gen Verbal_IPC = 1 if are15a1 != 1 & are15a2 != 1 & are15a3 != 1 & are15a4 != 1 & are15a5 == 1

Do you know how to include the OR | ?

[Nick Cox]

gen Verbal_IPC = min(are15a1, are15a2, are15a3, are15a4) > 1 & are15a5 == 1

See also https://journals.sagepub.com/doi/pdf...36867X19830921

[Belcaryn Catanza]

Thanks Nick... I'm back with a more complicated question.


I have to now determine Verbal ipc If ONLY ONE of the 4 variables are15a1, are15a2, are15a3, are15a4 is "Always or Often"

I have dichotomised these 4 variables to Yes 1 or No 0. But i cant figure out the formula to create a new variable if ONE of these four variables are a yes AND are15a5 is 1 (Never)

I thought the code you gave me would work but it just gives me 0 which doesn't look right

gen VERBAL_ipc = min(Verbal1, Verbal2, Verbal3, Verbal4) > 1 & are15a5 == 1.



So, I tried the following:


generate verbaladd = Verbal 1 + Verbal 2 + Verbal3 + Verbal 4

generate verbaladds = 1 if verbaladd > 1 & verbaladd !=.

This gives me a verbaladds variable which has 1 for a Always or Often on ANY of the four variables are15a1, are15a2, are15a3, are15a4

So I thought I could easily generate a VerbalIPCfinal variable by coding:

generate VerbalIPCfinal = 1 if verbaladds> 1 & verbaladds !=. & are15a5 == 1

But it just gives me all .'s


where am I going wrong??


Can I just change it in Excel and then import excel into STAT ready for the analysis? Would that be an issue if I did this?

[Nick Cox]

You can always use Excel if you prefer. What issues might there be? I don't use Excel for anything if I can avoid it, but I appreciate that many other people jump the other way. What I prefer, and many other people do so too, is being able to report if asked a complete script that goes all the way from original data to final results. Deciding that it is more convenient to do something in some other software is one thing. Reporting that you couldn't work out how to do it is another thing.

I have already given you several hints on useful functions and commands.

Your latest want is

ONLY ONE of the 4 variables are15a1, are15a2, are15a3, are15a4 is Always or Often AND are15a5 is Never

which can be approached like this:

egen verbal = anycount(are15a1 are15a2 are15a3 are15a4), values(4 5)
replace verbal = verbal == 1 & are15a5 == 1

I have some simple tips.

1. Sometimes it is evident that you can get there in one line. There is no shame, and much benefit, in breaking a calculation into simple steps. If there is a choice between being clear and seeming clever. go for being clear.

2. The Stata Journal paper linked in #8 covers various devices you're not using yet. Using if qualifiers may seem obvious but they aren't always the best tool towards clear and clean code.

3. egen is your friend. Did you look at the help to find out about relevant functions?

4. Make yourself a sandbox dataset with all 3125 outcomes and check that your indicators give exactly the right results after you've created them.

[Belcaryn Catanza]

Thanks for all the information Nick and the code, it works and I feel more confident to keep going with Stata.