Interesting. I would also ask, however, why you are using relative poverty and tax rates? Those depend not only on changes in RI, but changes elsewhere. (RI's relative poverty rate goes down if Texas' actually poverty rate goes up). I suspect that at least some of the relationship here is an artifact of these relative measures.

What happens when you plot the actual RI poverty rate against the lagged (or two-year average) RI tax rate? Isn't this what you're really interested in?

The relative indicators are to help disentangle what a state is actually doing for itself, versus what's happening because of national or international-factors beyond state government's control. It's hard to believe that the 10.2% poverty rate in RI at the height of the internet boom in 2000 (US poverty rate: 11.3%) has the same implications as the 10.3% rate in RI in 1994, when the US poverty rate was 14.5%. What was or wasn't RI doing to improve its situation around 2000, compared to 1994?

The relative numbers are also a cost-of-living/regional economic adjustment. Just making some numbers up, suppose taxes in RI in year 1 were 105% of the national average, while poverty was 70% of the national average. Then, over years 2, 3, and 4, the numbers averaged out to taxes at 108% of the national average and poverty at 50%. Maybe you have a good policy in that case – even though, compared to most other states, you didn't have much of a problem to begin with, while you raised taxes that were already "high".

We're not asking government to build perfection here, just to not make things worse than they were before.

Answer:
Rhode Island raised the minimum wage

Higher wages=less poverty

States that have the,much lower Federal minimum wage have the highest poverty rates
Low Minimum wage=more poverty

Leo, can you show some data for your hypothesis? First of all, the minimum wage is not a living wage and was not intended to be. Further, as the minimum wage went up in this state, so did the cost of living - especially housing. I would bet the cost of living outpaced the increase in the minimum wage and, therefore, the increase in the minimum wage was not a factor in the decrease in poverty.

If you've got data and analysis that demonstrates otherwise, bring it forward.

Andrew,

I agree with your impulse to try to factor out the effects of global influences on RI poverty rates to determine the relation of tax and poverty rates, but a) I don't think that can be done with just one data transformation, b) I don't think it can be done with any 2-variable scattergram and c) I'm pretty sure what you've done doesn't accomplish it. On the last point, try this for a check: If your approach succeeds in "controlling for" those global effects, you should find that the "RI relative poverty rate" is not correlated with the US poverty rate. If they are correlated, global effects still contaminate your measure. A quick pass through the data makes it appear that this is the case, because there is in fact a strong positive correlation between the two variables. I think you can verify this with a graphic, though a simple regression makes it clearer.

I'm not stating this more emphatically because I'm not sure your poverty data is the same as mine. I used the data at the link below, but I doubt it makes a difference.

http://www.census.gov/hhes/www/poverty/histpov/hstpov21.html ....

As I say above, your measure means that, even if the RI tax-to-poverty relation stays the same over time, your measure will cause it to appear to change if poverty levels or tax rates change in other states. That obviously makes it impossible to make statements about the RI relation.

In short, it is just not possible to draw the conclusion you want to from this 2-variable analysis.

There's more to be said...for instance that I don't agree at all that your "relative tax" measure makes sense, that it's not clear whether your poverty rates are "before transfer" or "after transfer", that it matters more how tax dollars are spent than how many are collected, etc.

I hope you will accept a friendly suggestion. If AR wants to do original empirical research and present conclusions based on data, you might find someone with stats training who can test out and look for trouble in your analyses. I think you will increase your chance of presenting more convincing analyses. (I'd offer my own services, but I'd charge you for it :

Thomas,

You use the term "empirical", yet the approach you are taking towards a correlation between variables normed to national averages that is stronger than a correlation between absolute variables is that you a-priori don't like the normed variables, so let's ignore them. That's not an "empirical" approach.

The statement that it is "impossible" to make statements about RI involving relative variables is dubious at best. This correlations present in this and the previous post exist. Unless someone is claiming that they are either a total coincidence, or that poverty rate relative to the national average and tax burden relative to the national average are basically the same quantity, they are just as worthy of explanation as any other correlations of the same strength.

Mobility is an active part of the debate about RI's taxation and poverty policies. Various hypotheses have been advanced that people seeking benefits will [or won't] move to places with more generous welfare policies, or that people will [or won't] move away from places with high taxes, relative to places with lower ones. Relative measures are valid -- even necessary -- to describe the environment that motivates and supports such movement.

More technically, I've made no claims of monocausality, nor that I expect RI's tax burden statistics by themselves to explain 100% of the variation in RI's poverty rates. When looking through multiple dependent variables (one choice being tax-burden) to explain the variation of an independent one, one general procedure is to begin with a variable that independently correlates well, then adding others. The fact that adding additional terms to a multiple regression's first term doesn't mean that that the first term correlation doesn't exit or isn't meaningful. Nor does the fact that there may be latent variables that better explain a causal relation between two quantities mean that the observable correlation doesn't really exist or isn't really meaningful. It is possible that it's not the tax burden that's the direct problem (it could be, for example, that RI manages of its human services programs very badly, and that is a major factor in the correlation), but the correlations oberved are too strong to allow "they don't matter" to be a valid response.

Andrew,

Here's a response. .Your statements are in italics

"You use the term "empirical", yet the approach you are taking towards a correlation between variables normed to national averages that is stronger than a correlation between absolute variables is that you a-priori don't like the normed variables, so let's ignore them. That's not an "empirical" approach."

Andrew, I'm sorry to say it, but the above statement is complete nonsense. I used the data to look at the correlation between the US poverty rate and the "RI relative poverty rate" . It is positive, fairly strong, and statistically significant. That correlation indicates that you have NOT successfully removed the effect of global influences on the RI poverty rate. I may be right and I may be wrong, but the conclusion is based on data, so In what way is that conclusion "a priori" and "not empirical"?

This correlations present in this and the previous post exist.

Yes, the correlations exist, but that doesn't necessarily tell us much, if anything. There's an (apocryphal?) story in the stats literature about a researcher who found a strong positive correlation between teacher salaries and champagne sales, and concluded that giving teachers raises was a waste because they would only spend it on booze. The correlation did in fact exist, but the conclusion was obviously nonsense. Once you control for the rise in real wages across all professions, there turns out to be no separate effect for teachers. Higher wages overall lead to higher teacher salaries and higher wages overall create more disposible income and higher liquor sales. There are a lot of zero-order correlations like this that disappear, or even reverse sign (go from positive to negative or the reverse), when you properly control for other variables.

Relative measures are valid -- even necessary -- to describe the environment that motivates and supports such movement.

I agree that they may be necessary, and they are valid if and only if they are correctly constructed. The question is whether you've constructed them correctly. I think you haven't.

When looking through multiple dependent variables (one choice being tax-burden) to explain the variation of an independent one, one general procedure is to begin with a variable that independently correlates well, then adding others.

Let's start by clearing up the concepts of "dependent" and "independent" variables. The "dependent" variable is the one that you're trying to explain. Its values are dependent on the values of the "explanatory" or "independent" variables. (I'll assume you just misspoke, as I often do when I write too fast… so this was for the benefit of the other readers)

The fact that adding additional terms to a multiple regression's first term doesn't mean that that the first term correlation doesn't exit or isn't meaningful.

The above sentence is missing a word or two, but I understand what you're saying. However, the statement is not really correct. Adding a second term to a multiple regression does indeed sometimes show that the first term is meaningless. See my comments on the teacher salary example above.

but the correlations observed are too strong to allow "they don't matter" to be a valid response.

Again, you don't know what a correlation means until you've controlled for other influences. I don't think they said "they don't matter". I think I said that it's not clear that they are meaningful.

The other thing going on in this data is that it's time-dependent, while your presentation is cross-sectional. The poverty rate in any one year probably depends, more than anything else, on the poverty rate in the previous year. You just can't move people into and out of poverty that fast, no matter how much you move factors that influence poverty rate. Without a term that measures the "year-1" effect of the dependent variable, I think it's highly likely that you have a serious autocorrelation problem (over time correlation in the error terms of the regression, for those who care). Ignoring this serial autocorrelation has led to many serious errors.

Statistics don't prove anything by themselves. They function as evidence in a causal argument, or (if you like) "story". Your causal story is that high (relative) tax rates "cause" (are positively associated with) high (relative) poverty rates. Your evidence is that high (relative) tax rates precede high (relative) poverty rates (in your model, a high relative tax rate in year Y-1 is associated with a high poverty rate in year Y),

OK,; but now try this. Let's flip the causal argument on it's head. Leaving aside all my objections to your relative measures for the moment, do a regression in which we predict the relative tax rate in year Y using the relative poverty rate<./i> in year Y-1 as the "independent" or "predictor" variable. Seriously…go ahead and do the regression. I have. Report what you find. You will find that poverty rate is as good predictor of tax rate as tax rate is a predictor of poverty rate. So, I can flip the story you tell, and say that tax rates respond to poverty rates rather than the reverse. Here's the story: RI raises its tax rate as poverty rates rise. That may be because a) need increases as poverty rises and/or b) tax revenues fall as poverty increases, so rates need to rise to compensate.