Chapter 4

The World of Numbers

There are three kinds of lies: lies, damned lies, and statistics.

Mark Twain

When trying to understand economic and social disparities, statistics are often used, both to convey the magnitude of those disparities and to try to establish their causes. To some, numbers may convey a sense of objective, hard facts. But, even when the numbers are correct, the words that describe what the numbers are measuring may be incorrect or misleading. These include such basic numbers as income, unemployment rates and rates of arrest.

Numbers may also be misleading, not because of any intrinsic defects in either the numbers themselves or in the words describing them, but because of implicit assumptions about the norms to which those numbers are being compared. Here the seemingly invincible fallacy of assuming an even or random distribution of outcomes as something to expect, in the absence of such complicating causes as genes or discrimination, can make many statistics that show very disparate outcomes be seen as indicating something fundamentally wrong in the real world, rather than something fundamentally wrong with the assumptions behind the norms to which those outcomes are being compared.

Neither logic nor empirical evidence provides a compelling reason for expecting either equal or random outcomes among individuals, groups, institutions or nations—or of natural phenomena such as tornadoes or earthquakes, for that matter.

When used with an awareness of their pitfalls, statistics can be enormously valuable in testing competing hypotheses about disparate outcomes. But statistics may nevertheless be grossly misleading when they are distorted by errors of omission or errors of commission.

ERRORS OF OMISSION

The mere omission of one crucial fact can turn accurate statistics into traps that lead to conclusions that would be demonstrably false if the full facts were known. This often happens in comparisons of different ethnic groups and different income classes, among other comparisons.

Group Disparities

In the course of a long and heated campaign in politics and in the media during the early twenty-first century, claiming that there was rampant discrimination against black home mortgage loan applicants, data from various sources were cited repeatedly, showing that black applicants for the most desirable kind of mortgage were turned down substantially more often than white applicants for those same mortgages.

In the year 2000, for example, data from the U.S. Commission on Civil Rights showed that 44.6 percent of black applicants were turned down for those mortgages, while only 22.3 percent of white applicants were turned down.1 These and similar statistics from other sources set off widespread denunciations of mortgage lenders, and demands that the government “do something” to stop rampant racial discrimination in mortgage lending institutions.

The very same report by the U.S. Commission on Civil Rights, which showed that blacks were turned down for conventional mortgages at twice the rate for whites, contained other statistics showing that whites were turned down for those same mortgages at a rate nearly twice that for “Asian Americans and Native Hawaiians.”

While the rejection rate for white applicants was 22.3 percent, the rejection rate for Asian Americans and Native Hawaiians was 12.4 percent.2 But such data seldom, if ever, saw the light of day in most newspapers or on most television news programs, for which the black-white difference was enough to convince journalists that racial bias was the reason.

That conclusion fit existing preconceptions, apparently eliminating a need to check whether it also fit the facts. This one crucial omission enabled the prevailing preconception to dominate discussions in politics, in the media and in much of academia.

One of the very few media outlets to even consider alternative explanations for the black-white statistical differences was the Atlanta Journal-Constitution, which showed that 52 percent of blacks had credit scores so low that they would qualify only for the less desirable subprime mortgages, as did 16 percent of whites. Accordingly, 49 percent of blacks in the data cited by the Atlanta Journal-Constitution ended up with subprime mortgages, as did 13 percent of whites and 10 percent of Asians.3 In short, the three groups’ respective rankings in terms of the kinds of mortgage loans they could get was similar to their respective rankings in average credit ratings.

But such statistics, so damaging to the prevailing preconception that intergroup differences in outcomes showed racial bias, were almost never mentioned in most of the mass media. With credit ratings being what they were, the statistics were consistent with Discrimination IA (judging each applicant as an individual), but were reported in the media, in politics and in academia as proof of Discrimination II, arbitrary bias against whole groups.

While the omitted statistics would have undermined the prevailing preconception that white lenders were biased against black applicants, that preconception at least seemed plausible, even if it failed to stand up under closer scrutiny. But the idea that white lenders would also be discriminating against white applicants, and in favor of Asian applicants, lacked even plausibility. What was equally implausible was that black-owned banks were discriminating against black applicants. But in fact black-owned banks turned down black applicants for home mortgage loans at a higher rate than did white-owned banks.4

Household Income Statistics

It is, unfortunately, not uncommon to omit statistics that are discordant with prevailing preconceptions. This has become a common practice in politics, in the media and even in much of academia. Such errors of omission are not confined to mortgage loan issues, but are also common in many discussions of income statistics.

Household income data, for example, are often used to indicate the magnitude of economic disparities in a society. But to say that the top 20 percent of households have X times as much income as the bottom 20 percent of households exaggerates the disparity between flesh-and-blood human beings, which can be quite different from disparities between income brackets. That is because, despite equal numbers of households in each 20 percent, there are far more people in the top 20 percent of households.

Census data from 2002 showed that there were 40 million people in the bottom 20 percent of households and 69 million people in the top 20 percent of households.5 Such facts are usually omitted in statistics about disparities in incomes.

No doubt people in the top quintile average higher incomes than people in the bottom quintile. But the fact that there were also 29 million more people in this top quintile exaggerates the disparity in incomes among people. Later data for 2015 from the U.S. Bureau of Labor Statistics indicated that there were now over 36 million more people in the top quintile than in the bottom quintile.6

Moreover, the number of people earning income was four times as large in the top quintile as in the bottom quintile.7 Most households in the bottom quintile have no one working.8 How surprising is it when four people working earn more income than one person working? That is yet another of the errors of omission, when the truth would undermine a prevailing preconception.

There are not only different numbers of people per household at different income levels, there are also different numbers of people per household from one ethnic group to another, and different numbers of people per household from one time period to another. Omitting those differences when drawing sweeping conclusions can distort the meaning or implications of those statistics.

As the Bureau of the Census pointed out, more than half a century ago, the number of American households has been increasing faster than the number of people.9 In short, American households tend to contain fewer persons per household over time—a trend continuing into the twenty-first century.10 There are not only smaller families in later times, more individuals are financially able to live in their own individual households, rather than live with relatives or roommates, or live as individual roomers or in boarding houses, as average incomes rise from generation to generation.

When income per person is rising over the same span of years when the average number of persons per household is declining, that can lead to statistics indicating that the average household income is falling, even if every individual in the country has a higher income.

For example, if per capita income rises by 25 percent over some span of years, during which the average number of persons per household declines from 6 persons to 4 persons, then four people in the later period have as much income as five people had in the earlier period. But that is still not as much as six people had in the earlier period, so average household income falls, statistically, even if everyone’s income has risen by 25 percent.

Household income statistics can be misleading in other ways. If two low-income people are sharing an apartment, in order to make the cost of rent less burdensome to each, and if either or both has an increase in income, that can lead to one tenant moving out to live alone in another apartment—and that, in turn, can lead to a fall in average household income.

If, for example, each of the two tenants had an income of $20,000 a year initially, and later both reach an income of $30,000 a year, leading to each living in a separate apartment afterwards, that will mean a fall in household income for these individuals from $40,000 a year to $30,000 a year. There will now be two low-income households instead of one, and each household will be poorer than the one they replaced. Again, a rise in individual income can be reflected statistically as a fall in household income.

Since most income is paid to individuals, rather than to households, and “individual” always means one person while “household” can mean any changeable number of persons, why would household income statistics be used so often instead of individual income statistics?

Clearly, omitting individual income statistics, and using household income statistics instead, is less useful to someone seeking the truth about economic differences among human beings. But household income statistics can be very useful for someone promoting political or ideological crusades, based on statistics that exaggerate income disparities among people.

For purposes of seeking the truth about the economic well-being of the American people, among the simplest and most straightforward statistics are data on per capita real income—that is, total money income divided by the number of people in the population, and adjusted for inflation. This is, however, a statistic seldom cited in income controversies, and seldom featured in official U.S. Bureau of the Census publications.

Time and Turnover

Among the factors often omitted, or distorted, in discussions of income disparities is the time dimension. People in the bottom 20 percent are often spoken of as “the poor” and, if the income in that quintile has not changed much over some span of years, it may be said that the income of “the poor” has stagnated. But the great majority of people initially in the bottom quintile do not stay there. There is nothing mysterious about the fact that most people tend to begin their working years in a lower occupation, with lower incomes than they will have in later years, after they have acquired more experience, skills and maturity, as well as a longer track record by which they can be judged.

A University of Michigan study that followed a given set of working Americans from 1975 to 1991 found that 95 percent of the people initially in the bottom 20 percent were no longer there at the end of that period. Moreover, 29 percent of those initially in the bottom quintile rose all the way to the top quintile, while only 5 percent still remained in the bottom 20 percent.11

Since 5 percent of 20 percent is one percent, only one percent of the total population sampled constituted “the poor” throughout the years studied. Statements about how the income of “the poor” fared during those years would apply only to that one percent of the people.

Similar distortions of reality occur when the time dimension is ignored in discussing people in the upper income brackets, who are often also spoken of as if they were a permanent class of people, rather than transients in those brackets, just like “the poor” in lower brackets. Thus a New York Times essay in 2017 referred to “This favored fifth at the top of the income distribution” as having collected “since 1979” a far greater amount of income than others.12

Considering how much turnover there was among people in different quintiles from 1975 to 1991, the implicit assumption that the same people were in the top quintile over the even longer period from 1979 to 2017 is a staggering assumption. But of course the very idea of turnover was omitted.

Another of the relatively few statistical studies that followed a given set of Americans over a span of years found a reality very different from what is usually portrayed in the media, in politics, or in academia: “At some point between the ages of 25 and 60, over three-quarters of the population will find themselves in the top 20 percent of the income distribution.”13

For most Americans in other quintiles to envy or resent those in the top quintile would mean envying or resenting themselves, as they will be in later years. What the New York Times chose to call a “favored fifth” is in fact a substantial majority of all Americans. Moreover, it seems doubtful that three-quarters of all Americans receive their income as a favor, rather than by working for it.

Calling people in particular income brackets “the poor” or “the rich” implicitly assumes that they are enduring residents in those brackets, when in fact most Americans do not stay in the same income quintile from one decade to the next.14 Similar patterns of transience in low-income brackets have been found by studies in Australia, Canada, Greece, Britain and New Zealand.15

The turnover rate among people in the highest income brackets is even greater than that of the population in general. Fewer than half of those Americans who were in the much-discussed “top one percent” in income in 1996 were still there in 2005.16 Although people in that bracket have been referred to as “the best-off one in one hundred,”17 that is true only as of a given instant. Over the course of a lifetime, the proportion of people in that bracket is one in nine, since 11 percent of Americans are in that bracket at some point in their lives.18 People initially in the top one hundredth of one percent had an even faster turnover, and those with the 400 highest incomes in the country turned over fastest of all.19

Crime Statistics and Arrest Statistics

Some of the most gross distortions of reality through errors of omission have involved quite simple omissions. No one needs to be an expert on the complexities of statistical analysis in order to see through many statistical fallacies, including those based on simple omissions. But it does require stopping to think about the numbers, instead of being swept along by a combination of rhetoric and statistics.

Statistics cited in support of claims that the police target blacks for arrests usually go no further than showing that the proportion of black people arrested greatly exceeds the roughly 13 percent of the American population who are black.

If anyone were to use similar reasoning to claim that National Basketball Association (NBA) referees were racially biased, because the proportion of fouls that referees call against black players in the NBA greatly exceeds 13 percent, anyone familiar with the NBA would immediately see the fallacy—because the proportion of black players in the NBA greatly exceeds the proportion of blacks in the American population.

Moreover, since blacks are especially over-represented among the star players in the NBA, the actual playing time of black players on the basketball court would be even more disproportionately higher, since it is the players on the court who get cited for fouls more so than secondary players sitting on the bench.

What would be relevant to testing the hypothesis that blacks are disproportionately targeted for arrest by the police, or disproportionately convicted and sentenced by courts, would be objective data on the proportions of particular violations of the law committed by blacks, compared to the proportions of blacks arrested, convicted and sentenced for those particular violations.

Such objective data are not always easy to come by, since data reflecting actions by the police would hardly be considered valid as a test of whether the actions of the police were warranted. However, there are some particular statistics that are both relevant and independent of the actions of the police.

The most reliable and objective crime statistics are statistics on homicides, since a dead body can hardly be ignored, regardless of the race of the victim. For as long as homicide statistics have been kept in the United States, the proportion of homicide victims who are black has been some multiple of the proportion of blacks in the population. Moreover, the vast majority of those homicide victims whose killers have been identified were killed by other blacks, just as most white homicide victims were killed by other whites.

Since the homicide rate among blacks is some multiple of the homicide rate among whites, it is hardly surprising that the arrest rate of blacks for homicide is also some multiple of the rate of homicide arrests among whites. What is relevant in such statistical comparisons is not the proportion of blacks in the general population, but the proportion of blacks among people who commit a particular crime.

Another violation of the law that can be tested and quantified, independently of the police, is driving in excess of highway speed limits. A study by independent researchers of nearly 40,000 drivers on the New Jersey Turnpike, using high-speed cameras and a radar gun, showed a higher proportion of black drivers than of white drivers who were speeding, especially at the higher speeds.20

This study, comparing the proportion of blacks stopped by state troopers for speeding with the proportion of blacks actually speeding, was not nearly as widely accepted, or even mentioned, by either the media or by politicians, as other studies comparing the number of blacks stopped by state troopers for speeding and other violations with the proportion of blacks in the population.21

Yet again, specific facts have been defeated by the implicit presumption that groups tend to be similar in what they do, so that large differences in outcomes are treated as surprising, if not sinister. But demographic differences alone are enough to lead to group differences in speeding violations, even aside from other social or cultural differences.

Younger people are more prone to speeding, and groups with a younger median age tend to have a higher proportion of their population in age brackets where speeding is more common. When different groups can differ in median age by a decade, or in some cases by two decades or more,22 there was never any reason to expect different groups to have the same proportion of their respective populations speeding, or to have the same outcomes in any number of other activities that are more common in some age brackets than in others.

The omission of data on the proportion of blacks—or any other racial group—engaged in a given violation of law, as distinguished from the proportion of blacks or others in the population at large, is sufficient to let racial profiling charges prevail politically, despite their inconsistency with either logic or evidence.

Some professional statisticians have refused to get involved in “racial profiling” issues because these issues are so politically charged. As a professor of criminology explained: “Good statisticians were throwing up their hands and saying, ‘This is one battle you’ll never win. I don’t want to be called a racist.’”23

Among the other consequences is that many law enforcement officials also see this as a politically unwinnable battle, and simply back off from vigorous law enforcement, the results of which could ruin their careers and their lives. The net result of the police backing off is often a rise in crime,24 of which law-abiding residents in black communities are the principal victims.

Some people may think that they are being kind to blacks by going along with unsubstantiated claims of racial bias and discrimination by the police and by the criminal justice system. But, as distinguished black scholar Sterling A. Brown said, long ago: “Kindness can kill as well as cruelty, and it can never take the place of genuine respect.”25

ERRORS OF COMMISSION

Statistical errors of commission include lumping together data on things that are fundamentally different, such as salaries and capital gains, producing numbers that are simply called “income.” But calling things by the same word does not make them the same things.

Other errors of commission include discussing statistical brackets as if they represented a given set of flesh-and-blood human beings called “the rich,” “the poor” or “the top one percent,” for example. Errors of commission also include using survey research to resolve factual issues that the inherent limitations of survey research make it unable to resolve.

Capital Gains

While annual income statistics for individuals avoid some of the problems of household income statistics, both of these sets of statistics count as income (1) annual wages, salaries and other incomes earned and paid during the same year and (2) income from capital gains accrued over some previous span of years, and then turned into cash income during a given year. Treating the incomes earned by some individuals over various numbers of years as being the same as incomes earned by other individuals in just one year is like failing to distinguish apples from oranges.

Capital gains take many forms from many very different kinds of transactions. These transactions range from sales of stocks and bonds that may have been bought over a span of prior years to sales of a home or business that has increased in value over the years.

If a farm was purchased for $100,000 and then, 20 years later—after the farmer has built barns and fences, and made other improvements to the land and the structures on it—the farm is sold for $300,000, that sale will result in a net increase of the owner’s income by $200,000 in the particular year when the farm is sold. Statistically, that $200,000 that was earned over a period of 20 years will be recorded the same as a $200,000 salary earned by someone else in just one year.

Looking back, that farmer has in reality earned an average of $10,000 a year for 20 years, as increases in the value of the farm, through the investment of time, work and money on the farm. Looking forward, the farmer cannot expect to earn another $200,000 the following year, as someone with a $200,000 annual salary can.

Capital gains in general are recorded in income statistics as being the same as an annual salary, when clearly they are not. Nor is there some easy formula available to render salaries and capital gains comparable, because capital gains by different individuals accrue for differing numbers of years before being turned into cash income in a given year.

If capital gains were equally present at all income levels—say, 10 percent of all incomes being capital gains—then the disparities in income statistics might not be affected as much. But, in reality, low annual incomes are far more likely to be salaries or wages, and very high annual incomes are far more likely to be capital gains. For example, someone making twenty thousand dollars a year is probably getting that from a pay check, while someone making twenty million dollars a year is more likely to be making that much money from capital gains of one sort or another.

The exceptionally high rates of turnover of people at very high income levels reinforce this conclusion. Internal Revenue Service data show that half the people who earned over a million dollars a year, at some time during the years from 1999 through 2007, did so just once in those nine years.26

This does not imply that all the others in that bracket made a million dollars every year. Another study, also based on tax data, showed that, among Americans with the 400 highest incomes in the country, fewer than 13 percent were in that very high bracket more than twice during the years from 1992 to 2000.27 The highest incomes are usually very transient incomes, reinforcing the conclusion that these are transient capital gains rather than enduring salaries.

All of this distorts the implications of income statistics that treat annual earnings and multi-year capital gains as if they were the same. Talk of how much of a country’s income is received by the top ten percent, or by the top 400, proceeds as if this is a given set of people. But, because of the high turnover rate in high income brackets, there can be thousands of people in the “top 400” during just one decade.

During the period from 1992 to 2014, for example, there were 4,584 people who were in the top 400 income earners, according to Internal Revenue Service data. Of these, 3,262 were in that bracket just one time during those 23 years.28 When incomes received by thousands of people are reported statistically as if these were incomes received by hundreds of people, that is a severalfold exaggeration of income disparities—in this case, more than a tenfold exaggeration.

Such data are also relevant to the oft-repeated claim that “the system is rigged” by the wealthy. That claim certainly fits the prevailing social vision. But if the question is whether it also fits the facts, then it can be tested like any other hypothesis. In light of these data, if the top 400 income recipients in the United States rigged the system, and 71 percent of the people in the “top 400” are in that category just one time during a period of more than two decades, why would anyone rig the system in such a way that they themselves would be unlikely to remain in the topmost income bracket? It would have to be some of the most incompetent rigging imaginable.

It is a similar story as regards the 400 richest people in the world, who had net losses of $19 billion in 2015.29 As of 2016, the number of billionaires in the world was slightly fewer than in 2015, while the total wealth held by all the world’s billionaires declined by $570 billion on net balance.30 If the world’s richest people had in fact “rigged the system,” surely they could have done better for themselves than that.

Racial and Ethnic Disparities

In trying to determine the reasons for economic and social disparities between blacks and whites, some observers attribute these differences primarily to policies and practices by people outside the black community, while other observers attribute these same differences to internal differences in behavior between black and white Americans.

In seeking to resolve this issue, sociologist William Julius Wilson relied heavily on statistics from opinion surveys. These surveys, according to Professor Wilson, show that “nearly all ghetto residents, whether employed or not, support the norms of the work ethic.”33 In one survey, “fewer than 3 percent of the black respondents from ghetto poverty census tracts denied the importance of plain hard work for getting ahead in society, and 66 percent expressed the view that it is very important.”34

After admitting that “surveys are not the best way to get at underlying attitudes and values,”35 Professor Wilson nevertheless presented—as a refutation of “media perceptions of ‘underclass’ values and attitudes” in inner-city ghettos—the fact that “residents in inner-city ghetto neighborhoods actually verbally endorse, rather than undermine, the basic American values concerning individual initiative.”36

Despite Professor Wilson’s reliance on opinion surveys to refute claims that ghetto residents have different cultural values from those of the American population as a whole, there is no necessary correlation between what people say and what they do. A survey of low-income people by Columbia University researchers showed that 59 percent regarded buying goods on credit as a bad idea. Nevertheless “most of the families do use credit when buying major durables.”37

Economists tend to rely on “revealed preference” rather than verbal statements. That is, what people do reveals what their values are, better than what they say. Even when people give honest answers, expressing what they sincerely believe, some people’s conception of hard work, for example, need not coincide with other people’s conception, even when both use the same words.

When black students in affluent Shaker Heights spent less time on their school work than their white high school classmates did, and spent more time watching television,38 that was their revealed preference. Data from other sources show even greater differences between the time devoted to school work by black Americans and by Asian Americans in high school.39 Nor are such differences peculiar to blacks or to the United States. In Australia, for example, Chinese students spent more than twice as much time on their homework as white students did.40

How surprised should we be that Asian students in general tend to do better academically than white students in general, in predominantly white societies such as Australia, Britain or the United States? The same pattern can be seen among whole nations, as such Asian countries as Japan, South Korea and Singapore likewise show patterns of hard work by their students and academic results on international tests that place these countries above most Western nations.41

Statistics compiled from what people express verbally may be worse than useless, if they lead to a belief that such numbers convey a reality that can be relied on for serious decision-making about social policies.

Incidentally, the high correlation between the amount of work that different groups put into their education and the quality of their educational outcomes does not bode well for theories of genetic determinism. When we find some race whose students spend less time and effort on their school work than students in some other race, and yet get educational results superior to the results of hard-working students in that other race, this would be evidence supporting the genetic hypothesis. But such evidence does not seem to be available.

When trying to determine, from statistics, how much discrimination—in the sense of Discrimination II—is involved in income differences between groups, one of the common errors of commission that can make such comparisons unreliable is to compare individuals with supposedly the “same” education or other qualifications, when the qualifications are in fact not the same in any meaningful sense. Comparing blacks and whites with the “same” number of years of schooling was obviously not a comparison of blacks and whites with the same education during much of the early twentieth century.

Most blacks were concentrated in the South during that era, and their racially segregated schools usually had shorter school years, among other disparities.42 According to a study of that era, “A black pupil attending school three months of the year for six years would have finished at most half the grades completed by a white attending six months for six years.”43 Statistically, however, they both had the “same” number of years of education.

Clearly there was Discrimination II in such situations, but where the discrimination took place cannot be determined by where the statistics were collected. That is, when black and white workers with the “same” amount of schooling received different pay, that was not necessarily all due to employers’ biased discrimination, as distinguished from biased discrimination that took place for years in the school system, before black workers reached an employer.

Similarly in a later era, after schools were no longer racially segregated by law, but black high school graduates had scores on education tests that were lower than the scores of white students who were years younger.44 Here the failure of employers to pay black workers what white workers with the “same” education were being paid need not be due to Discrimination II in the workplace, even if the statistics were collected at the workplace. But, when comparing individuals with the same results on mental tests, blacks and whites received comparable pay.45

Statistics often do not have nearly as much detail as would be necessary to compare people who were truly comparable. A very similar problem can affect comparisons of male and female workers with the “same” qualifications in gross terms. But when a higher proportion of women are part-time workers, and they are workers with fewer years of continuous employment experience (due to taking time out to care for small children), or with college degrees in subjects that do not prepare them for careers in high-paying professions, comparisons of women and men with the “same” qualifications in gross terms are comparisons of apples and oranges.

In situations where data are available on a wider range of work qualifications, so that truly comparable individuals from different groups can be compared, income disparities tend to shrink, often to the vanishing point, and sometimes the inequality reverses.46 As far back as academic year 1972–73, for example, white faculty members had higher incomes than black faculty members. But, when comparing those faculty members with Ph.D.s in the same fields, from similarly high-ranked departments in their respective disciplines, and with similar numbers of published articles, black faculty members had higher incomes than their white counterparts.47 Similarly, male faculty members in general had higher incomes than female faculty members in general. But, among similarly qualified faculty, women who never married earned higher incomes than men who never married.48

Minimum Wages and Unemployment

One of the important areas in which survey research has done major damage has been in trying to resolve differences of opinion as to the effect of minimum wage laws on unemployment. Advocates of minimum wage laws argue that such laws raise the incomes of the poor, while critics argue that these laws cause more of the poor to be unemployed, because low-income workers tend to be workers with few skills and/or little work experience, so that employers find them worth employing only at low wage rates. Despite an abundance of detailed statistics on unemployment, this controversy has raged for generations.

Part of the problem is that, as we have seen in other contexts, most of what are called “the poor” are not permanent residents in low-income brackets, any more than other people are permanent residents in other income brackets. About half of all Americans earning at or near the minimum wage rate are from 16 to 24 years of age,49 and of course they do not remain young permanently. So, when people say, as Senator Ted Kennedy once said, “Minimum wage workers have waited almost 10 long years for an increase,”50 they are not talking about a given set of human beings, but about a statistical category containing an ever-changing mix of people.

Because young people are usually, almost by definition, less experienced as workers, their value to a prospective employer tends to be less than the value of more experienced workers in the same line of work. Some young people may acquire valuable work skills through education, but education also takes time, and people grow older with the passage of that time.

Often what younger, inexperienced workers acquire from an entry-level job is primarily the habit of showing up every day and on time, and the habit of following instructions and getting along with others. After having acquired work experience in some simple, entry-level job, most young beginners go on to other jobs, with different employers, where work experience of some sort may be a prerequisite for getting hired. Simple as such things may seem, the absence of a work experience prerequisite can negate whatever other good qualities a young worker may have, but which have not yet had a chance to manifest themselves in a work situation.

High rates of employee turnover, sometimes exceeding 100 percent per year, are common in many entry-level jobs in retail businesses or fast-food restaurants.51 These jobs are stepping stones to other jobs with other employers, though some observers falsely call entry-level jobs “dead-end jobs.”52 If workers in fact stayed on permanently in such jobs, which usually have no automatic promotions ladder, those workers would in fact be in dead-end jobs. But, when the average tenure of supermarket employees has been found to be 97 days, that is clearly not the case.53

Like most things in a market economy, inexperienced and unskilled workers are more in demand at a lower price than at a higher price. Minimum wage laws, based on what third parties would like to see workers paid, rather than being based on those workers’ productivity, can price unskilled workers out of a job. This traditional economic analysis has been challenged by advocates of minimum wage laws, and survey research data have been a major part of that challenge.

Back in 1945, Professor Richard A. Lester of Princeton University sent out questionnaires to employers, asking how they would respond to higher labor costs. Their responses, which were not along the lines of traditional economic analysis, convinced Professor Lester that the traditional economic analysis was either incorrect or not applicable to minimum wage laws.54 However, what traditional economic analysis seeks to do is predict economic outcomes, not predict how people who are surveyed will answer questionnaires. Moreover, outcomes are not just the fruition of beliefs or intentions, as we have seen in discussions of the costs of discrimination.

Decades after Professor Lester’s challenge to traditional economic analysis, other economists, also at Princeton, again challenged traditional economic analysis on the basis of survey research, though now by surveying the same employers before and after a minimum wage increase, and asking each time how many employees they had. The answers convinced the Princeton economists that the minimum wage increase had not reduced employment. They and their supporters therefore declared the traditional analysis to be a “myth” that had now been “refuted.”55

Devastating criticisms of the Princeton economists’ conclusions were made by some other economists, who challenged both the accuracy of their statistics and the logic of their conclusions.56 But, even if the Princeton economists’ statistics were completely accurate, that would still not address the key weakness of survey research in general—which is that you can only survey survivors. And what may be true of survivors need not be true of others in the same circumstances who did not survive in those particular circumstances.

An extreme hypothetical example may illustrate the point that is applicable in less extreme situations. If you wished to determine empirically whether playing Russian roulette was dangerous, and did so through survey research, you might send out questionnaires to all individuals known to have played Russian roulette, asking them for information as to their outcomes.

After the questionnaires were returned and the answers tabulated, the conclusion from these statistics might well be that no one was harmed at all, judging by the answers on the questionnaires that were returned. Not all questionnaires would have been returned, but that is not uncommon in survey research. Basing your conclusions on the statistical data from this research, you might well conclude that you had “refuted” the “myth” that playing Russian roulette was dangerous. This is the kind of result you can get when you can only survey survivors.57

In the case of minimum wage studies, if all the firms in an industry were identical, then any reduced employment resulting from the imposition of a minimum wage, or the raising of an existing minimum wage rate, would appear as a reduction of employment in all the firms. But, in the more usual case, where some firms in a given industry are quite profitable, others are less profitable and still others are struggling to survive, unemployment resulting from a minimum wage can push some struggling firms out of the industry—and reduce the number of their replacements, now that labor costs are higher and profits are more problematical.

The only firms that can be surveyed for their employment data, both before and after the minimum wage was imposed or raised, are the firms that were there in both time periods—that is, the survivors. If there has been a net decrease in the number of firms, the employment in these surviving firms need not have gone down at all, regardless of a decline in employment in the industry as a whole. The firms surveyed are like the people who survived playing Russian roulette, which may well be a majority in both cases, though not an indicative majority in terms of the issue at hand.

Empirically, a study of the effect of raising local minimum wage rates on employment in restaurants in the San Francisco Bay Area found that the principal effect was through some restaurants going out of business—and reducing the number of new firms entering to replace them. Those restaurants going out of business were primarily restaurants rated lower in quality. Employment in five-star restaurants was unaffected.58 In Seattle as well, the response to a local minimum wage rate increase was that a number of restaurants simply closed down.59 These now non-existent restaurants obviously cannot be surveyed.

The amount of labor demanded can be measured either by the number of workers employed or by the number of hours that they work, or both. A study published by the National Bureau of Economic Research measured employment by hours of work, as well as by the number of workers employed, and concluded that “the minimum wage ordinance lowered low-wage employees’ earnings by an average of $125 per month in 2016.”60 Thus a theoretical increase in income from a higher minimum wage rate became, in the real world, a significant decrease in income. That is, even workers who kept their jobs worked fewer hours and therefore earned less money.

Another problem with trying to determine the effect of a minimum wage law on unemployment is that the proportion of the work force directly affected by a minimum wage is often small. Therefore unemployment among that fraction of the work force can be swamped by the normal fluctuations in the unemployment rate among the larger number of other employees around them.

This is less of a problem in situations where most of the employees are earning a wage low enough to be directly affected by a minimum wage law. But five-star restaurants were unlikely to be having inexperienced teenagers delivering food to their upscale customers’ tables, even if restaurants like McDonald’s or Burger King often have teenagers delivering food over the counter to their customers.

Other ways of assessing the effect of a minimum wage on unemployment would include gathering data restricted to just the kinds of inexperienced and unskilled workers directly affected, such as teenagers. We have already seen, in Chapter 2, how minimum wage laws affect both the teenage unemployment rate in general and racial disparities in teenage unemployment rates as well.

Yet another way of assessing the effect of minimum wage laws on unemployment would be to gather unemployment data on places and times where there have been no minimum wage laws at all, so that these unemployment rates could be compared to unemployment rates in places and times where there have been minimum wages laws—especially where these have been comparable societies or, ideally, the very same society in the same era, with and without a minimum wage law.

By focusing on teenagers in general, or black teenagers in particular, it is possible to see the effects of minimum wage laws more clearly and precisely, since these are workers on whom such laws have their greatest impact, because these are a population most lacking in education, job skills and experience, and therefore earning especially low wage rates. Moreover, there are extensive statistics on what happened to these populations in the labor markets from the late 1940s to the present.

As we have seen, what is most striking about statistics on American teenage unemployment rates in the late 1940s is that (1) these unemployment rates were only a fraction of the levels of unemployment to which we have become accustomed to seeing in later decades, and (2) there was little or no difference between the unemployment rates of black and white teenagers then.

Internationally, unemployment rates have been markedly lower in times and places where neither governments nor labor unions set most wage rates. Most modern nations have had minimum wage laws, but the few that have not have tended to have strikingly lower unemployment rates. These would include Switzerland and Singapore today, and Hong Kong under British rule, prior to the 1997 return of Hong Kong to China. There was also no national minimum wage law in the United States before the Davis-Bacon Act of 1931, which impacted wage rates in the construction industry.

As for hard data on unemployment rates in these places and times, The Economist magazine reported in 2003: “Switzerland’s unemployment neared a five-year high of 3.9% in February.”61 But this “high” (for Switzerland) unemployment rate returned to a more normal (for Switzerland) 3.1 percent in later years.62

In 2013, Singapore’s unemployment rate was 2.1 percent.63 In 1991, when Hong Kong was still a British colony, it too had no minimum wage law, and its unemployment rate was under 2 percent.64 In the United States, the last administration with no national minimum wage law at any time was the Coolidge administration in the 1920s. During President Coolidge’s last four years in office, the annual unemployment rate ranged from a high of 4.2 percent to a low of 1.8 percent.65

Yet discussions of minimum wage laws, even by some academic scholars, are often based on the intentions and presumed effects of these laws, rather than being based on empirical evidence as to their actual consequences.

IMPLICATIONS

The emphasis on complex statistical analysis in economics and other fields—however valuable, or even vital, such statistical analysis may be in many cases—can lead to overlooking simple but fundamental questions as to whether the numbers on which these complex analyses are based are in fact measuring what they seem to be measuring, or claim to be measuring. “Income” statistics which lump together annual salaries and multi-year capital gains are just one of many sets of statistics which could stand much closer scrutiny at this fundamental level—especially if laws and policies affecting millions of human beings are to be based on statistical conclusions.

What can be disconcerting, if not painful, are the simple and obvious fallacies that can pass muster in intellectual circles when these fallacies seem to advance the prevailing vision of what is called “social justice.” Among prominent current examples is French economist Thomas Piketty’s large international statistical study of income inequality, which was instantly acclaimed in many countries, despite such obvious and fundamental misstatements as one pointed out by Professor Steven Pinker of Harvard:

In addition to speaking of percentages as if they represented a given amount of income or wealth over the course of a century, Professor Piketty also made such assertions as that, in income, “the upper decile is truly a world unto itself,”67 when in fact just over half of all Americans are in that upper decile at some point in their lives.68 When Piketty said that the top one percent sit atop the “hierarchy” and “structure of inequality,”69 he again verbally transformed a changing mix of people in particular income brackets into a fixed structure rather than a fluid process, in which most Americans do not remain in the same quintile from one decade to the next.

Such misstatements are different expressions of the same fundamental misconception. As an empirical study of the 400 richest Americans pointed out, Piketty “naively assumes that it’s the same people getting richer.”70 But the majority of the 400 richest Americans have earned their fortunes in their own lifetimes, rather than being heirs of the 400 largest fortunes of the past.71

Such misconceptions are not peculiar to Professor Piketty. Nor are these the only problems with his statistics. But that such simple and obvious misstatements can pass muster in intellectual circles is a problem and a danger that goes far beyond Thomas Piketty.

Whether income differences are measured before taxes or after taxes can change the degree of inequality. If inequalities are measured both after taxes and after government transfers, whether in money or in goods and services, that can reduce the inequality considerably, when high-income people pay higher taxes and low-income people receive most of the government transfers.

Statistics on tax rates themselves can be grossly misleading when changes in tax rates are described in such terms as “a $300 billion increase in taxes” or “a $300 billion decrease in taxes.” In reality, all that the government can do is change the tax rate. How much tax revenue that will produce depends on how people react. There have been times when higher tax rates have produced lower tax revenues, and other times when lower tax rates have produced higher tax revenues,72 as well as times when tax rates and tax revenues moved in the same direction.

During the 1920s, for example, the tax rate on the highest income Americans was reduced from 73 percent to 24 percent—and the income tax revenue rose substantially73—especially income tax revenues received from people in the highest income brackets. Under the older and much higher tax rate, vast sums of money from wealthy investors were sheltered in tax-exempt securities, such as municipal bonds. The total amount of money invested in tax-free securities was estimated to be three times the size of the annual budget of the federal government, and more than half as large as the national debt.74

Such vast and legally untaxable sums of money caught the attention and aroused the ire of Secretary of the Treasury Andrew Mellon, who declared it “repugnant” in a democracy that there should be “a class in the community which cannot be reached for tax purposes.”75 Failing to get Congress to take steps toward ending tax exemptions for incomes from particular securities,76 Secretary Mellon sought instead to lower the tax rates to the point where it would in fact lead to collection of more tax revenues.

Tax-exempt securities tend not to pay as high a rate of return on investments as other securities, whose earnings are taxed. It made sense for wealthy investors to accept these lower rates of return from tax-exempt securities when the tax rate was 73 percent, but not after the tax rate was lowered to 24 percent. In terms of words on paper, the official tax rate on the highest incomes was cut from 73 percent to 24 percent in the 1920s. But, in terms of events in the real world, the tax rate actually paid—on staggering sums of money previously untouchable in tax shelters—rose from zero percent to 24 percent. This produced huge increases in tax revenues received from high-income people, both absolutely and as a percentage of all income taxes collected.77

This increase in income taxes collected from high-income taxpayers was a result of the plain fact that 24 percent of something is larger than 73 percent of nothing. Tax rate cuts in some later administrations also led to increases in tax revenues.78 For example, a front-page news story in the New York Times of July 9, 2006 said: “An unexpectedly steep rise in tax revenues from corporations and the wealthy is driving down the projected budget deficit this year.”79

However unexpected this increase in tax revenues may have been to the New York Times and others decrying “tax cuts for the rich,” this was precisely the kind of outcome predicted and expected by others in various administrations over the years, who had urged that tax rates be cut, in order to get money disgorged from tax shelters and invested in the market economy. This included people in the Coolidge, Kennedy, Reagan and George W. Bush administrations, where there were similar outcomes.80 But the very possibility that tax rates and tax revenues can move in opposite directions is seldom mentioned in the media—a crucial error of omission.

These are not simply arguments about history. Among the consequences in our own time is that proposals to reduce income tax rates are automatically met with objections to reducing income tax revenues. In the Wall Street Journal of January 31, 2018, for example, economist Alan Blinder objected to tax rate cuts on grounds that “the deficit is already too large.”81

This is in defiance of what the New York Times reported about the unexpected reduction of the deficit by increased tax revenues during the administration of President George W. Bush. It is also in defiance of a record-breaking budget surplus after tax rates were reduced in the 1920s—a surplus large enough to allow about one-fourth of the national debt to be paid off.82 Like many others, Professor Blinder proceeded as if it were axiomatic that tax rate reductions mean tax revenue reductions.

There is, of course, no guarantee of what any given tax rate reduction will lead to in a given set of circumstances. But Professor Blinder’s assertion was not based on any argument that a tax rate reduction under particular current circumstances would lead to a reduction in tax revenues. There was in fact no argument whatever on that point, nor apparently any sense of need to make such an argument. Similarly, a twenty-first century book on President Calvin Coolidge likewise asserted that, as a result of the tax rate cuts during his administration, “the bounty that the rich enjoyed sapped the U.S. Treasury of funds it might have used for other ends.”83 Thus a record-breaking budget surplus under President Coolidge was verbally transmuted into a deprivation of funds, with the turn of a phrase.

All the voluminous and detailed statistics on tax rates and tax revenues published by the Internal Revenue Service, going back more than a hundred years, might as well not exist, as far as many of those with the prevailing social vision are concerned. This is ultimately not a question about history, but about what such heedlessness implies for the present and still more so for the future.