Santorum blames gay marriage for bad economy

[Wisconsin+17]

You must be logged in to read this quote.

Not saying it isn't, or that you are pulling something shady

You must be logged in to read this quote.

This is how I would do it if I were doing it myself, and then overlay them on top of each other.

This would be fantastic.

[Wisconsin+17]

Alcon, let me guess - you trained in social sciences?

[Alcon]

But, understanding that we have a disagreement on the relative virtues of the first versus second method, are you ready for me to run the first method?  Although it lacks trendlines, it is certainly the less volatile of the two methods, and we know we won't have to throw it out due to appearance of noise.  If you don't have any suggested tweaks, shall I run it?

(Why I think this less-volatile method is better: Even if this method does not use trendlines, it would still include any "hit" marriage took in MA even if the effect was delayed...so even under the delayed-effect hypothesis, the "hit" would show up in the less-volatile method's numbers.  That, combined with the lesser volatility, is why I'll argue it's the superior method.)

Alcon, let me guess - you trained in social sciences?

My major is in Political Science but it's more humanities-focused.  This is just the method of analyzing public policy that makes most sense to me.  It actually involves real-world application (unlike some more straightforward, concrete stats stuff) and it isn't made-up bullsh**t (unlike a lot of policy argumentation.)  I think stats + ethical theory is the best hope for making policy in a way that doesn't just exclusively involve knee-jerk/ideology-driven power-plays.

What's your field of study?  (Not to be presumptuous, but the way you write suggests you have some formal education in this or something similar-ish)

[Wisconsin+17]

Well, my stats training is in Physics.

Feel free to run the analysis with the years together if we can do both of them.

I'm expecting that with the other method that we'll see convergence.

[Alcon]

Ahh, cool Yeah, same vocabulary, very different sort of Stats (from what I've gathered)

Here's the low-variance "post vs. pre" method:

Change in divorce rate, post-2004 period versus pre-2004 period
Massachusetts: -9.8%
United States: -13.4%
Positive change (Massachusetts minus U.S.): -3.6%

Change in marriage rate, post-2004 period versus pre-2004 period
Massachusetts: -7.4%
United States: -17.6%
Positive change (Massachusetts versus U.S.): +10.2%

Change in marriage:divorce ratio, post-2004 period versus pre-2004 period
Massachusetts: +2.7%
United States: -3.9%
Positive change (Massachuestts versus U.S.): +6.6%

Yep, so:  Massachusetts's decline in divorce rate was a bit lower than the U.S.'s between these two periods.  However, Massachusetts saw much less of a decline in the marriage rate than the U.S. did.  Consequentially, Massachusetts actually saw an increase in marriages-per-divorce between the two periods (+2.7%) while the other states saw an average decline (-3.9%).

It's a small difference (these are percents of decimals), but using this measurement of divorce and marriage rates, Massachusetts pretty clearly fared better than the rest of the U.S. overall after legalizing gay marriage.  Massachusetts "wins" this method.

[Wisconsin+17]

You must be logged in to read this quote.

Ok, now what would happen if you toss out the high MA and the low MA numbers? The US numbers should be ok. Tossing out high low is how we 'correct' for the variance in the MA numbers.

[Alcon]

So, how do you want to do the trendline bit?  You're right that it's best to avoid using one year as a big data point, so I'm reluctant to just set 2003 or 2004 as year 0.  How about I average 1999-2003 and set that as year 0, make 2004 year 1, 2005 year 2, and so on?  That creates a model that is extremely sensitive to trends post-2003, for better or for worse, but mitigates the variance problem a little.

I'm using the marriage:divorce ratio for this, if that's cool.  That seems like the best metric, although I suppose I can also use (marriage rate-divorce rate).  Whichever you like...

[Alcon]

You must be logged in to read this quote.

Ok, now what would happen if you toss out the high MA and the low MA numbers? The US numbers should be ok. Tossing out high low is how we 'correct' for the variance in the MA numbers.

Well, that's how you control for outliers, but it jacks up the variance since you reduce the number of years from 7 to 5.  Mixed blessing...and the MA "outliers" are pretty tame compared to many of the smaller states.

But pretty much nothing happens:

Change in divorce rate, post-2004 period versus pre-2004 period
Massachusetts: -9.9%
United States: -13.4%
Positive change (Massachusetts minus U.S.): -3.5%

Change in marriage rate, post-2004 period versus pre-2004 period
Massachusetts: -8.2%
United States: -17.6%
Positive change (Massachusetts versus U.S.): +9.4%

Change in marriage:divorce ratio, post-2004 period versus pre-2004 period
Massachusetts: +1.8%
United States: -3.9%
Positive change (Massachuestts versus U.S.): +5.7%

There's just no rational statistical method where MA doesn't fare better post-2004 than the other states.

[Wisconsin+17]

I was thinking about this. It might be more useful to chart the slopes rather than the absolute values, and then plot all the slopes of all 50 states on top of them.

That should give us a pretty good approximation as to what is going on with each states. It should show up as to whether or not MA is an outlier.

[Wisconsin+17]

You must be logged in to read this quote.

Now we have a much more reliable sample. We haven't gotten to the point yet where we can say that this isn't due to random factors.

What's the standard deviation of the 100 averages (pre + post)?

[Alcon]

I was thinking about this. It might be more useful to chart the slopes rather than the absolute values, and then plot all the slopes of all 50 states on top of them.

That should give us a pretty good approximation as to what is going on with each states. It should show up as to whether or not MA is an outlier.

Yep, that's what I was planning to do.  I just need to know the timeframe and the statistic to chart, like I asked above.

(Also, like I alluded to previously, I excluded 8 states that have years missing...shouldn't matter, but just want to be transparent.)

You must be logged in to read this quote.

Now we have a much more reliable sample. We haven't gotten to the point yet where we can say that this isn't due to random factors.

What's the standard deviation of the 100 averages (pre + post)?

I honestly think this is all probably noise.  I really doubt legalizing gay marriage has enough effect on marriage and divorce rates overall, which are dominated by heterosexuals, to outpace the effect of noise.

The Excel STDEV function is screwy, but it looks like it's 7.6%.

Mean is -3.9%.

Range is +17.0% (Iowa) to -20.3% (Nevada)

Distribution looks a lot like a bell curve, with Massachusetts on the happy side of the middle (14th of 43.)

I'd never suggest that this is anything like conclusive evidence that gay marriage is good for "marital health" of states, but this metric says it's much likelier to be good than bad, if it's either.

[Wisconsin+17]

No need to do the other analysis. We're getting good data here. If it's a bell curve, I'm happy with it.

[Wisconsin+17]

You must be logged in to read this quote.

That's just about what we'd expect from a random sample. Sorry. There's nothing to indicate that there's any structural reason for the change beyond noise.

I thought it might be Iowa + Nevada. Interesting.

Thanks for doing all this. Sorry we got nothing definitive out of it.

[Alcon]

All right.

For what it's worth, I did a simple CORRELATION (correlation between year and the change in the marriage:divorce ratio) in Excel, which would get about the same result as the slope method.  Again, Massachusetts appears to be on the happier side of the middle of the pack, 20th of 43.  The marriage:divorce ratio didn't fare particularly well during this period, which isn't a big surprise with the economic downturn.  However, only 3 states reached statistically significant downward correlations -- Virginia, Wisconsin and Arkansas.  All three of these states banned gay marriage during this period although I'm not reading much of anything into that.

So, where are we at with this debate?  Before, you were saying the "broken window" hypothesis is probabilistic based on secondary correlations.  Now, we've isolated the variable as well as we can.  I did one test that is high-variance, high-trend-sensitivity; I did another that is mid-variance, mid-trend-sensitivity.  Both of these are direct, instead of secondary correlations.  They both indicate that the state with gay marriage (Massachusetts) has fared better in marital health by all metrics than states without gay marriage.  Eyeballing these numbers, it appears states that banned gay marriage also fared worse than those that didn't.  In fact, I bet if I run a correlation between gay marriage opposition and marital health deterioration between this period, it might reach statistical significance.

No matter, though.  The point is:

1. There's no evidence that gay marriage deteriorated the "marital health" of Massachusetts versus the states without gay marriage.

2. If anything, Massachusetts' "marital health" relatively improved.

3. This new evidence is more direct than the data you were using to bolster the "broken window" hypothesis, which you made the empirical cornerstone of your argument.

Are you ready to reevaluate?

[Wisconsin+17]

Interesting. Wisconsin, Virginia and Arkansas. Not the ones I would have expected to see. 

You must be logged in to read this quote.

Yes, but we've also shown that this isn't statistically significant.

You must be logged in to read this quote.

Well, what I'm seeing is that marriage rates in general are declining across America. This reinforces my opinion (and Santorum's), that this is an overarching negative trend. 

Two, the rise in the ratio of marriage/divorce that we see in MA seems to be ephemeral. It took the lowest divorce rates recorded for any state in the US to bring that number to about par for the US. If we see another year like the last one, I suspect that MA will actually trend negative in that ratio.

So no, there's nothing here that would convince me to re-evaluate my position.

[Wisconsin+17]

Maybe this could be published somewhere? I'm sure someone would be interested in this.

[Alcon]

...Broken window, yeah sure maybe man.  But...you don't see anything here that suggests that you lack empirical evidence for your assertion that The Gays are contributing to that "broken" window?  That was...the entire point of this exercise.

How many states have to legalize gay marriage and then perform better-than-average in "marital health" trends before you change your mind on gay marriage...?

[Wisconsin+17]

You must be logged in to read this quote.

If you showed me marriage rates going up - I'd see things differently. This was actually a really depressing study.
 

You must be logged in to read this quote.

Statistically significant is the point here. I'm not sure why I should give weight to something that hasn't been shown to be statistically significant. Like I said, the best roll of the dice in 2006 still doesn't pull it up to the point where it would be statistically significant.

I think we can conclude from here on that the picture isn't going to improve.

Honestly, I thought your conclusion was rubbish as soon as I saw the actual data series for MA. That pretty much sealed it for me. Sorry.

[Alcon]

Wow, dude.

The only way that you can accept gay marriage is if it reverses, all by itself, the downward trend in "marital health"?

You don't even care if the variable "presence of gay marriage" seems to have no negative effect on "marital health"?  Because Massachusett's superior "marital health" trend was unarguably probabilistic, statistically significant in most interpretations and a direct, not secondary, correlation.  It's better, harder evidence than you were previously using to ground your argument...and now suddenly you're discounting it?  When before you were all "they may be indirect, man, but look at the empirics"?  Maybe I misunderstand something.  If not, how the hell can you defend that...?

But, yeah, go ahead and explain to me what conclusion I made that's "rubbish."  This will be interesting.

sorry indeed...

[Comrade Funk]

Wait a minute: Nixon is sane!?!

Yes he was. He was just one of the biggest weasels in history.

[Wisconsin+17]

You must be logged in to read this quote.

That, or it shows to be statistically significant in this survey here. It wasn't, so yeah, I'm unconvinced that this is anything more than noise. You said it so yourself.

You must be logged in to read this quote.

Spin, spin, spin, spin, spin, spin.

Sorry, no. It's not statistically significant. It's about 1 standard deviation off the norm, and when you've got 50 states, and MA is about at the 1/3rd level, yeah, ok.

Given a truly random sample, how many would we expect to see at 1 standard deviation? 1/3rd? Gosh, that was easy.

Even if this were statistically significant, you would have to explain why the other 20 or so states with a greater improvement than MA are significant as well, and account for MA.

See what I'm saying here? That it isn't statistically significant just says, ok, so it's up, but it's not up enough to make a difference from what we could see randomly.

You see this all the time. I was asked to do a survey of boy/girl ratios at birth. In the UK you'll see some isolated high ratios, but when you have that many samples, you would expect to see outliers at a specific rate. Which is what we see here.

I bet if we made the cutoff '07 this behaviour disappears.

[Alcon]

That, or it shows to be statistically significant in this survey here. It wasn't, so yeah, I'm unconvinced that this is anything more than noise. You said it so yourself.

That's the same thing!  You're believing your hypothesis because of a secondary correlation, when isolating the variable indicates that there's no evidence that gay marriage is the explanatory variable!  How does that work?  My entire point was that the evidence for your hypothesis sucks, and your presumed explanatory variable (presence of gay marriage contributing to a "broken window") is unsubstantiated when you isolate the variable.  Basically, you've set up a mode of analysis where you can stake a causal hypothesis and the only way to reject it is to prove that the subject of the causal agent (gay marriage) is so strong it actually reverses all other trends (deceased "marital health"), not just that there's no evidence it contributes to those trends when you isolate the hypothesized causal agent.  How can this make any sense to you?  How can you care about the secondary correlation, but not about isolating the variable you're trying to test for?  That is not how proving a hypothesis works in any branch of science because it is totally irrational.

Re: statistical significance: I was referring to the pre- vs. post-2004 change of MA vs. other states as being statistically significant, but I may have run the test wrong.  But are you accusing me of intentionally mis-running a statistical significance test?  Does that match my behavior in this thread at all?  It's like you've reverted to hackery once I ran this analysis  (Also, again, statistical significance is arbitrary.  94% certainty is still pretty damn likely, and is again, superior to your secondary correlation which can't even be tested for statistical significance but yet you accept your hypothesis as gospel truth anyway wtf.)

Either I'm misunderstanding it (I don't think so) or your methodology of analysis is almost completely illogical.

[Alcon]

Seriously, not to beat up on you, but I can't believe you're accusing me of arguing to your own conclusions, when your mode of analysis is built upon arbitrarily accepting objectively inferior evidence that matches your conclusion (secondary correlation with no statistical significance test possible) and arbitrarily accepting hypotheses until you are 95% confident that they're untrue...

Would you actually apply this mode of analysis to anything in your discipline?

[Torie]

How are we coming with the empirical evidence?  What is the reader's digest version? I have a little question. Why are hetero families in the most disarray where the specter of gay marriage is the least visible and distracting, or whatever the theory is?  I mean, how many gays are running around Mississippi agitating for gay marriage?  And why are black families in the most disarray?  I mean how much do most of those folks think about gay marriage at all, or even gays, except to disdain them perhaps?  They should be the least susceptible to the family toxic gay influence no?  Sorry to butt in.

[Alcon]

How are we coming with the empirical evidence?  What is the reader's digest version? I have a little question. Why are hetero families in the most disarray where the specter of gay marriage is the least visible and distracting, or whatever the theory is?  I mean, how many gays are running around Mississippi agitating for gay marriage?  And why are black families in the most disarray?  I mean how much do most of those folks think about gay marriage at all, or even gays, except to disdain them perhaps?  They should be the least susceptible to the family toxic gay influence no?  Sorry to butt in.

Ben's hypothesis is that gay marriage contributes to a "broken window effect" that is responsible for the decline of the institution of marriage in recent years.

His evidence is that "marital health" (divorce and marriage statistics) have not fared well nationally since gay marriage became legalized in Massachusetts and became a viable national issue.

I attempt to isolate the effect of gay marriage by comparing the "marital health" statistics in Massachusetts versus in the nation.  They failed to substantiate the hypothesis; although there is high variance, Massachusetts fared better by all metrics than the nation as a whole.

After I made a best-faith effort to isolate this variable and it did not substantiate his hypothesis (the opposite, if anything), he complained that Massachusetts' better performance could be statistical noise.

Basically, he's accepted the hypothesis that gay marriage contributes to the "broken window" of marriage based on a secondary correlation ("marital health" continues to decline as gay marriage was legalized.)  He dismisses any attempt to isolate the variable "presence of same-sex marriage" to figure out if it's an explanatory variable, unless it's an explanatory variable in the opposite direction, to statistical significance.

In other words, unless I can prove to statistical significance that gay marriage alone reverses the downward trend in marital health he's presuming that gay marriage contributes to the "broken window."  This is despite the fact that his only evidence for this is the national marital trends, and since he's not even isolating a variable in his analysis, he can't even perform a statistical significance test on that.

Really short version: He's using inferior, more general evidence to justify his hypothesis, and then arbitrarily rejecting more precise evidence.  It's absolute crap, and I don't really believe anyone sees reality that way unless they're arguing to conclusions.

Sorry, that's a little complicated and wordy, but we're talking about hypotheses, statistical analysis results, and statistical analysis methods at the same time, which tends to make for long sentences