Google+ Followers

Thursday, January 22, 2015

Inequality Models are Not Facts

The New York Times ran an article --  not an opinion piece, but an article -- last week entitled "Study Finds Local Taxes Hit Lower Wage Earners Harder".  It is a superficial summary of a report that is published every few years by a progressive think tank, The Institute on Taxation and Economic Policy ("ITEP"), which the article laughably describes as "nonpartisan".  The institute's board of directors includes Robert Reich, former Clinton Administration Secretary of Labor and a Berkeley academic, Robert Kuttner of The American Prospect, a leading progressive periodical and website, and other well-known progressives or associates of other progressive organizations.  The ITEP staff bios show that, with few exceptions, they have spent their entire adult lives inside the beltway or in surrounding states, beginning, in the main, with getting a degree (in what is usually not specified), at Georgetown or GWU, or perhaps as far away as UVa or U of Md.  Their bios cite prior contributions to the New York Times and the New Republic, but not a single publication of a different ideological stripe.  The organization's recent other publications include "Undocumented Immigrants' State and Local Tax Contributions", "State Tax Codes as Poverty Fighting Tools", "The Sorry State of Corporate Taxes" and "Options for Progressive Tax Relief".  It states on its website that it partners frequently with Citizens for Tax Justice, another, better-known progressive think tank that endorses Bernie Sanders' legislative proposals.  So this report is obviously partisan, unless as a New York Times writer or reader you are so steeped in progressive ideology that you can't see the rest of the world clearly. 

The report "Who Pays?  A Distributional Analysis of the Tax Systems of all 50 States" is available on the ITEP website. The proposition of the report is, as one might expect, that any tax system other than income taxes with progressive rates is unfair.  Most of the report is devoted to state-by-state tables that layout ITEP's calculations of tax burdens by income quintile, with the predictable breakdown of the top quintile into three subsets - top 1%, top 5% and the rest.  The report ranks each state in a tax inequality index of ITEP's own devising, in which four states without income taxes - WA, FL, TX and SD--  get ranked most unequal while states with high top rates on income -- OR, CA and DE -- get ranked least unequal.  Because DE has no sales tax at all, that combines with its "modestly progressive" income tax to make it the least unequal in the eyes of the ITEP staff.

The naive reader might think such a "study" simply consists of researchers going through tax return data, pulling out data on state and local taxes paid, then measuring those payments against income and putting the resulting ratios into a table.  The facts are the facts and the author of this post is just upset that progressives have found them out.  Ah, but "fact", dear reader, in social science data crunching is often the creation of the data cruncher, and that is the point of this post.  I consider myself a proud member of the "reality-based community" among which I counted many fellow citizens of the left back during the Bush-Cheney regime.  Sadly, when a progressive President came into power due to the unprecedented incompetence of said regime, the "reality-based community" got a lot smaller and more conservative.  The myths these days are confected on the left.  In this post, I will show how the data is manipulated to achieve a counter-factual depiction of tax payment distribution that comports with the progressive mindset of the organization that did the study. I don't claim that the manipulation stems from an outright intent to deceive; it is equally possible that it stems from the simple cognitive bias that can build up when people of like minds and backgrounds talk only to people of like minds and backgrounds.

The report does not detail the methodology of the computations, except indirectly in passing references to different kinds of taxes.  Sources for data were only partly disclosed in the report.  References are made to an underlying model, and I hunted down a short methodological note about it on ITEP's website that answered some of my questions.  The model appears to be based on 1980s and 1990s era academic research, probably reflecting when ITEP opened for business, with ad hoc updates, mostly of data sets, since then.  From these sources, I was able to identify at least some of the allocational and methodological decisions they made to generate their results.

For example, they claim to exclude all persons over 65 from the report because, to paraphrase page 19 of the report, "they get too many tax breaks" and their inclusion would obscure the picture ITEP wants to present of "the majority".  By this logic, all inequality studies can exclude "the 1%" or even the 1% and the two lowest quintiles, because the study would still be presenting a picture of "the majority". So one example of a choice made to slice the data to get to the desired result.
Secondly, they don't count all taxes.  They admit to excluding severance taxes on oil and gas extraction, which exist in most states and are certainly a major factor in Alaska.  They also exclude at least a portion of tourism taxes (as I write, I cannot recall if it is part or all but it's not that large a number either way).  The argument for both of those exclusions is that they are "exported" out of the state that imposes them, i.e. paid by its non-residents.  Which is true but they have to land somewhere and the vast majority of them land within the rest of the United States.  The model does not appear to include that second-stage allocation. 

Third, and a much larger contribution toward reaching a depiction of inequality, they selectively define what is counted as income.  From an endnote, it likely comes from IRS data which is well-understood to exclude many items, such as employer-provided health coverage to middle and upper middle class workers. But even for people in the lower quintiles of income, there are significant government benefits they receive that are not counted as income by the IRS.  For example, when a Medicaid patient sees a doctor and does not pay the real cost of the visit, that subsidy does not show up as the patient's income. If one lives in a rent-controlled apartment, or an affordable-housing complex where the landlord has agreed to a certain rent, even though there is not a formal rent control, that is an economic benefit that does not show up in one's tax return. Mass-transit subsidies are not acknowledged as income, etc. 

Similarly, a low-income household that sends children to public school does not recognize income based on a per child share of the education expenditures made by their school district for the benefit of those children,  But educational expenses make up the vast bulk of local government spending, and are the main use of property taxes!  In my town, school taxes make up 2/3 of my property tax bill, even though I have no children still attending the public schools.   There is a net transfer, not captured in traditional income categories, from taxpayers with a high income-to-child ratio to those with a low income-to-child ratio.  Government's role as middleman, or post-office, obscures that transfer, and the sponsors of inequality arguments, always people on the left, are disinclined to pierce that obscuring. 
ITEP defends this exclusion by saying, in substance, "yeah, well, an educated workforce benefits employers too."  But given that business profit margins are well-documented, and education-to-salary correlations have been extensively studied, that objection seems weak and convenient.  Employers don't capture 100% of the benefit of educating the workforce, given that they pay that workforce, you know, wages.  They capture at most a small fraction of it in the portion of their profit margin that may be allocated to worker productivity.  Thus, the portion of education spending captured by the student vs the portion theoretically captured by a future employer is as estimable as anything else ITEP chooses to estimate. I suspect they dodge this challenge because, by itself, it would kill their conclusion, given the growth of public sector education expenditure that has occurred in recent decades. 

Income earned in the "informal economy" usually is not included as income in these studies, although it obviously exists and is heavily weighted toward the lower and middle quintiles.  ITEP claims to include an estimate for it, but I am very suspicious of it, since none of the examples they give of informal income -- farm income? seriously? -- include the kind of off-the-books wages or payments to independent contractors that is all I have ever seen and I have seen dozens of examples of it.  For example, one of my pro bono clients was a maintenance worker in a Manhattan office building.  He made about $30,000 - $40,000 in salary and his wife received a small amount of disability payments.  That is not usually enough to live on in NYC so I asked what other resources he was living on.  He told me they got by due to a sideline of his doing exterminations in the outer boroughs on off-hours and weekends,  He told me the business had generated roughly $30,000/year for several years -- which he had never declared to the IRS (which had never audited him).   As he was in no position to pay back taxes, interest or penalties, he chose not to proceed with the personal bankruptcy filing, in which he would have had to disclose the income or perjure himself, and tried to tough it out on his own.

Or there was the retired NYS teacher who tutored one of my kids for two years in math but demanded payment in cash (despite the fact that NYS public sector workers have their pensions  exempt from state and local taxation). Or the guy who gave music lessons, or the unlicensed child care provider, the small job home repairs contractor, etc.

There is academic research that attempts to quantify the distribution of non-compliance with the Internal Revenue Code that supports the idea that high earners cheat at least as much as low earners.  It has been observed that given higher marginal tax rates the high earner has a greater incentive to cheat..  That observation, however, tends to overlook that, due to the plethora of income-based subsidies provided by state and federal governments, which phase out as one earns more, the highest marginal tax rates in the economy are in the working class which then supports the proposition that the working class has the highest propensity to cheat.  Also, such studies, in addition to the inherent sampling error, have data collection questions:  how does one know the IRS caught everybody and every item of noncompliance?  It's harder to detect small amounts of cheating and the auditing and collection effort may not be worthwhile. An agent may buy a questionable explanation if the amount in dispute is small.  Whereas a bigger target justifies a bigger effort, although admittedly they may have greater resources to challenge the IRS.  Further the noncompliance studies contain their own sets of assumptions and judgments, so citing them to support a second generation of allocation propositions compounds the risk of error.

There are other types of income that don't make it into the IRS form usually -- debt that the creditor fails to collect, but also doesn't formally release being one category of billions of dollars a year that goes unrecognized. 

Fourth and also highly significant in terms of the ultimate re-distribution of tax payments by the study's authors, the ITEP study allocates a significant portion of business and property tax payments to persons who are not the owners of the property or business. The justification is stated that many of those taxes are passed through. This is the concept of "tax incidence".  I found a paper on their website that says (in a section titled "Incidence Assumptions") that 1/2 of property tax charged on rented residential real property was passed through by their model to tenants -- clearly an estimate and otherwise I did not see a detailed justification of the 50% number. 

This is perhaps the best example of how one can create facts in a model.  Obviously, the person who gets the bill and writes the check to discharge taxes on rental property is the owner.  So that is the "fact" that exists in the data sampled.  Further, in accounting, for centuries, expense and income are supposed to be matched to make for an accurate and not misleading presentation, so there too the recipient of rental property income would be required to show the property tax as an expense of renting that property.  So based on those objective facts, one would assign all the property tax to the owner of property which would depict a more progressive distribution.

The concept of tax incidence changes that.  It posits that, whatever the paperwork and money flows look like, economically, the burden of a tax may fall on a person different than the one paying it.  In this case, it may fall on the tenant, on the theory -- note that this is a theoretical allocation, not a reality-based allocation  -- that the landlord charges enough rent to cover its expenses which include the property tax, thus some portion of the property tax is included in the rent.

Tax incidence can also be invoked in the context of corporate income taxes (it is posited, for example, that workers may bear the burden of corporate income taxes, in the form of reduced wages, because the employer has to make enough money to stay in business and the tax is an expense that competes with other expenses, like wages) and even in the context of sales taxes (when you buy an item that costs $100 and the state adds 5%, you are obviously paying $105 total to take the item with you, so would you have paid $105 to the store for the item?  In that case, the store owner is bearing the burden of the tax, not the consumer).  The ITEP model does not seem to re-allocate any tax other than property tax on residential real property however. I am not sure intellectually how one selects that tax for a counter-factual reallocation, and not others.  A landlord pays income tax on rental income too -- by the logic used to reallocate property tax, one would reallocate the rental income too.  A logical explanation, however, is the progressive authors of the study have a desire to push states toward a more income-tax dependent regime and thus would not want to reallocate income taxes away from high earners, regardless of intellectual consistency.

Whether any particular claim of tax incidence is true is difficult to prove.  From a common-sense point of view, even a vacant building has a property tax, so there is no simple pass-through going on.  I happen to own two condos that I rent out to the extent I can.  The property tax is the same whether I rent them out or not; they are just taxed as residential units.  Am I passing that through when I elect to rent the units out?  Or is the tax simply eating into my profit, because the tenant would pay $X rent regardless of how much property tax I have to pay?  I know of no way to prove either proposition objectively. 

Then there are further distributional propositions one could make: perhaps in good school districts, with high property taxes, a landlord can pass taxes through, because tenants will sacrifice to get into such a district, but logically in worse school districts, with a poorer population, a landlord may succeed in passing less taxes through (you can't collect what they can't pay).  None of this is discussed in the ITEP papers available on its website.   Really the 50% is plucked out of the air.  It's no more a "fact" than 25% or 75% would have been.

In the main, the extent of tax incidence of any particular tax is a conjecture derived from theoretical models that rest on debatable assumptions (like full employment and efficient labor markets in the arguments that workers bear the burden of corporate income taxes). The authorities ITEP cites in support of their Incidence Assumptions are just theoretical arguments based on propositions about elasticity of inputs and so forth that themselves are partly conjectures.  There is no broad, accepted dataset that one goes to and says what the tax incidence of some tax is. There is just no way to prove the extent of tax incidence across every tax on every property in every state, but that doesn't stop people from asserting it when it suits their purposes.

The incidence of property taxes has been the subject of academic conjecture for at least 4 decades, yet there are virtually no empirical examinations.  The only two I can find in the last 25 years that I can commend are one from 1994 by Carroll and Yinger in Boston, who concluded that only 15% of any property tax is passed through to tenants, and a large 20-year study recently concluded in Germany where it was concluded that, in the short-run, landlords absorb all or substantially all of any property tax increase, and are only able, over the long term, to pass through to tenants some of it where and when rental housing demand exceeds rental housing supply.  So there is no support I can find for anything close to the allocation made to tenants in the ITEP study, which is obviously a very significant part of their portrayal of a picture of inequality. 

And as a closing point about the biased nature of the ITEP model, please note how its disparate allocation of property taxes vs its disregard of the services typically paid for with property tax revenue, i.e., education.  As pointed out above, the model allocates zero income to households for schooling children.  Yet it allocates to renters 50% of the landlord's property taxes.  That is a sure way to make the distribution look regressive: allocate costs but not benefits.  Unfortunately, that is at odds with the most basic principles of financial presentation: match the person's or entity's income and expenses.  Here, the organization deliberately chooses to separate them, which is a big driver of the portrait they create.

A fifth example of putting a thumb on the scale is the reduction of the tax burden for higher earners for the federal deductibility of state income and property taxes, without including a similar offset for the lowest quintiles by virtue of the standard deduction they can take on federal income taxes.  Some portion of the standard deduction should logically be viewed as a substitute for state and local tax deduction.  This is a relatively minor factor in the ultimate redistribution though.

A final observation about the choices made in the model is not about accuracy or bias but about values, as President Obama might say.  The model allocates excise taxes on gasoline, alcohol and tobacco primarily to lower-income quintiles, which forms a substantial part of their tax burden in many states.  The alcohol and gasoline allocations are debatable I think, as wealthier families may have more cars per household, and may imbibe more expensive alcohol.  But in any case, it's hard to see why taxes on these items shouldn't be maintained, if not increased.  Inequality is not the only issue in America, although sometimes it seems to be the only one to progressives. 

So the overall point is this: models do not generate facts.  Models generate representations of reality. Representations have a tendency to reflect the predisposition of the person making them. Models usually contain assumptions and similar propositions that are not proven empirically and may be selective to suit a particular outcome, as real estate appraisals in the subprime crisis were.   Models may omit or manipulate data for a variety of reasons (my favorite "tell" is to search a paper containing an economic or similar model for variations on the word "simplify" which usually leads right to the things the model's author cut out of the data).  One need not debate whether the manipulation of data is in bad faith or reflects simple cognitive bias, but one must be aware of the risk that one or the other is present.