Voting Analysis, redux
candid told me I should have done the voting data analysis as a logistic regression, which was an excellent point. So I spent some of my "copious free time" re-learning how that works, and then re-parsed the NES survey data.
Population: Everyone in the NES survey who gave a valid responsive answer to every question pertaining to political affiliation and demographics data. I then culled out everyone who expressed no preference between the two major parties. N=823
The model tries to estimate P(R), the probability that a given voter in this population expressed a preference, to whatever degree, for the Republican party versus the Democratic party.
Variables found to be highly significant, and their weights:
INCOME +1.1 (this is a continuous variable from 1-23. The 1.1 represents the delta between bottom and top)
NHNCP -0.6 (non-hispanic with at least one parent who is not a U.S. citizen)
AGE<25 -0.6
POSTGRAD -0.8 (has a master's or PhD. MDs and lawyers did not fit the pattern)
OWNHOME +0.5 (indicated that they owned their home rather than rented)
WORKCLAS -0.5 (when asked if they self-identified as working class or middle class, chose the former)
WOMAN -0.8
RURAL +0.6 (from the possibilities rural,small town,suburb,big city,inner city)
SUBURB +0.7
PRAY12 +0.5 (when asked how often they prayed, chose one of the top 2 responses)
BIBLE1 +0.4 (believes the Bible is the literal word of God)
BIBLE3 -1.1 (believes the Bible was just written by men)
CATHOLIC -0.7 (self-identified as Catholic)
JEWISH -1.1 (self-identified as Jewish)
HISP -0.8 (self-identified as Hispanic)
BLACK -3.4 (self-identified as Black)
To put these numbers into some perspective, -3.4 is a GIGANTIC factor. All other variables being equal, it is enough to take someone from a 50% probability to 3%. A -0.7 swing, all things equal, changes someone from 50% to 33%
Variables that I tried in various forms, but weren't significant: Old age, being married, having kids, military service, unemployment status, retired status, treating income in any way other than a continuous scale, education other than POSTGRAD, any other way of considering urbanicity, religiosity, citizenship, or ethnicity
Population: Everyone in the NES survey who gave a valid responsive answer to every question pertaining to political affiliation and demographics data. I then culled out everyone who expressed no preference between the two major parties. N=823
The model tries to estimate P(R), the probability that a given voter in this population expressed a preference, to whatever degree, for the Republican party versus the Democratic party.
Variables found to be highly significant, and their weights:
INCOME +1.1 (this is a continuous variable from 1-23. The 1.1 represents the delta between bottom and top)
NHNCP -0.6 (non-hispanic with at least one parent who is not a U.S. citizen)
AGE<25 -0.6
POSTGRAD -0.8 (has a master's or PhD. MDs and lawyers did not fit the pattern)
OWNHOME +0.5 (indicated that they owned their home rather than rented)
WORKCLAS -0.5 (when asked if they self-identified as working class or middle class, chose the former)
WOMAN -0.8
RURAL +0.6 (from the possibilities rural,small town,suburb,big city,inner city)
SUBURB +0.7
PRAY12 +0.5 (when asked how often they prayed, chose one of the top 2 responses)
BIBLE1 +0.4 (believes the Bible is the literal word of God)
BIBLE3 -1.1 (believes the Bible was just written by men)
CATHOLIC -0.7 (self-identified as Catholic)
JEWISH -1.1 (self-identified as Jewish)
HISP -0.8 (self-identified as Hispanic)
BLACK -3.4 (self-identified as Black)
To put these numbers into some perspective, -3.4 is a GIGANTIC factor. All other variables being equal, it is enough to take someone from a 50% probability to 3%. A -0.7 swing, all things equal, changes someone from 50% to 33%
Variables that I tried in various forms, but weren't significant: Old age, being married, having kids, military service, unemployment status, retired status, treating income in any way other than a continuous scale, education other than POSTGRAD, any other way of considering urbanicity, religiosity, citizenship, or ethnicity