Logit regression on sub samples

[Raz Naji]

Hi,

I am currently trying to see what the effect of reading the Washington Times and the Washington Post has on the US presidential voting outcome on those who identify as conservatives and those who identify as liberals. My sample has either received the post, the times or nothing (control). My voting outcome is a dummy variable equal to 1 if they voted for democrats. My post and times variables are also dummy variables. I am splitting my sample into those who identify as liberal (conservative<4 where conservative is a scale from 1-7, 1 being 'extreme liberal' and 7 being 'extreme conservative') and conservative (conservative>4).

I am controlling for many variables including gender, age, preference and whether they have voted before.
Here is what my regression looks like so far:

logit voteddem2 post times Bfemale reportedage voted2001 voted2002 voted2004 Bpreferdem Bpreferrepub Bprefernoone if conservative>4
logit voteddem2 post times Bfemale reportedage voted2001 voted2002 voted2004 Bpreferdem Bpreferrepub Bprefernoone if conservative<4

Would this be a good way to measure the effectiveness of receiving the post/times on voting Republicans or Democrats in each sample?

Thanks!

[Carlo Lazzaro]

Raz:
welcome to this forum.
Some comments about your post:
1) your -of- clauses seems inconsistent, because the value=4 is excluded.
I would have expected:

 logit voteddem2 post times Bfemale reportedage voted2001 voted2002 voted2004 Bpreferdem Bpreferrepub Bprefernoone if conservative>4 logit voteddem2 post times Bfemale reportedage voted2001 voted2002 voted2004 Bpreferdem Bpreferrepub Bprefernoone if conservative<=4

2) you would be better off with creating a new two-lecel categorical variable, say -new_conservative-:

gen new_conservative=0 if conservative<=4
replace new_conservative if conservative> 4 & conservative!=.
label define new_conservative 0 "Conservative" 1 "Progressive"
label val new_conservative new_conservative

[Raz Naji]

Hi,
Thank you for your response!

I was initially planning to run conservative=4 as some robustness check for my regression. However your code does work, but can I check if this would be creating a dummy variable for conservative and liberal?

I also have a question on how I can see the effect of reading the Washington Times on liberals on voting for Republicans instead of Democrats (voteddem2 = 0). Is this possible?

Best,

Raz

[Raz Naji]

Raz:
welcome to this forum.
Some comments about your post:
1) your -of- clauses seems inconsistent, because the value=4 is excluded.
I would have expected:

logit voteddem2 post times Bfemale reportedage voted2001 voted2002 voted2004 Bpreferdem Bpreferrepub Bprefernoone if conservative>4 logit voteddem2 post times Bfemale reportedage voted2001 voted2002 voted2004 Bpreferdem Bpreferrepub Bprefernoone if conservative<=4

2) you would be better off with creating a new two-lecel categorical variable, say -new_conservative-:

gen new_conservative=0 if conservative<=4
replace new_conservative if conservative> 4 & conservative!=.
label define new_conservative 0 "Conservative" 1 "Progressive"
label val new_conservative new_conservative

Hi,
Thank you for your response!

I was initially planning to run conservative=4 as some robustness check for my regression. However your code does work, but can I check if this would be creating a dummy variable for conservative and liberal?

I also have a question on how I can see the effect of reading the Washington Times on liberals on voting for Republicans instead of Democrats (voteddem2 = 0). Is this possible?

Best,

Raz

[David Radwin]

I can't say for sure, but my guess is that conservative == 4 indicates "neither conservative nor liberal" identification. (Political scientists are fond of 7-point Likert scales with a neutral middle category.)

If so, you might want to use different code to make a dummy variable where the value is missing if conservative == 4, such as replacing the second line of Carlo's example with

replace new_conservative if conservative> 4 & conservative!=. & conservative!=4

As for whether reading a particular newspaper causes people to vote for a candidate of a particular party, you usually cannot make that inference from observational data. The reasoning is too long and off-topic to elaborate here. For a brief explanation, though, please see question 1 of a midterm exam for an introductory political science methods course for which I was a teaching assistant ages ago. It's on pp. 1-2 of http://davidradwin.com/ps3/Midterm1_answers.pdf.

[Carlo Lazzaro]

David is right and I've overlooked the indifference point of the Likert scale.