Rank the Different Components /

¶ … rank the different components / larger areas they discuss, then proceed to smaller focus some specific areas of question, then point out ways the study could be improved. The critical approach will start from global, macroscopic overall description, and focus in to describe the continents on that globe or the different major sections of the paper, through details that comprise the broader map. Micro-level focus on details will pay particular attention to specific notes, where many of the assumptions underlying the conclusions this paper draws, originate. I compare this publication with similar papers by parallel Congressional and Executive agencies and a few academic sources. Finally areas for improvement and further research will reflect implications and information necessary or useful but omitted.

Review of the Paper in General

A critical review points out strengths and weaknesses, and there are numerous factors supporting the validity and relevance of this paper. Strengths include that it was published by the Congressional Budget Office (CBO), i.e. The U.S. Congress, so the imputed authors have political stake in publishing accurate information at least; we hope their individual biases will be offset through the peer review embodied in that institution. The CBO has published numerous papers on topics similar enough for comparison. This paper includes prominent but heterodox economists like Paul Krugman and Thomas Lemieux, which improves ethos compared to examples we will consider briefly later. There are many other strengths we will pose alongside weaknesses in our closer focus below as well.

We must limit our expectations of this document stringently: This February 2011 publication "documents changes in the level and distribution of hourly wages received by workers in the United States between 1979 and 2009" (CBO 2011, unnumbered Preface), and defines some of the major supply and demand factors that affect wages for different skill sets; productivity; gender; education levels and other characteristics. Keep the limiting terms "distribution" and "wages" foremost. We are also told the paper will consider the role of "labor market institutions" that most affect wages, which we will also consider closely. This all occurs under the CBO's usual disclaimer foreswearing any type of recommendation derived by them from this "objective" and "impartial" study (Preface).

The global conclusions of this paper are that inflation-adjusted wages at the median have increased over the period 1979-2009, by about 20%; that median wage was about $17 at period's end; the gap between both high and low wages have moved away from the 50th percentile for both men and women for different reasons all around, and that these changes took different courses over time. This evidence has been discovered analyzing wage earnings that are limited in specific ways, adjusted for inflation under specific constructions, organized into percentiles, and weighted for specific factors I analyze below. These data describe events affected by market conditions, most importantly technology and productivity innovation effects; global trade patterns, and immigration, which are separate and distinct from institutional factors particularly minimum wage levels and union representation. We must consider these continents more specifically in order to describe the globe just described.

Data Set Considerations

The practice of ranking data into different grades, describing the middle and the two tails in reference to that 50th percentile has become so standardized over the last 100 years we will leave criticism of the model aside as beyond the scope of this paper. The data set is large enough; it came from the U.S. Census and was thus probably normal and random, and outliers have been removed by considering the tenth and 90th percentiles rather than the absolute tails, which is strong. Downside outliers were already removed by the minimum wage, to large degree. We consider median instead of mean, which begs discussion somewhere else, but which helps overcome distortions in distribution, which is the purview of this study, and this too is all robust.

This is about as far as we can get critiquing this paper without discovering opportunities for potential controversy. This publication relies on assumptions that begin with the adjustment of data for comparability, so-called 'chaining' both in the indexing for inflation and in weighting wages for hours worked. Nor are the U.S. Census Current Population Survey data immune from criticism on grounds of subject reporting errors, as evidenced by sustained attack by partisans for the Current Employment Survey, which boils down to administrative vs. survey data in general (Gould 2003). Imputing earnings for missing answers is a subordinate potential confound that has relatively small but real effects. The biggest problems with the data set however in my opinion, arise through chaining wages for hours worked and some specific questionable decisions made selecting data for consideration, and assigning wage rates for non-wage payroll or excluding employment classes that probably bear significantly on both the validity, relevance and thus utility of this report.

Some of these potential disagreements can be described if not solved, and thus dispensed with, in less space. The dispute over survey data as compared to administrative data, is deep but fairly clear cut. The Census Current Population Survey data, on which this survey is based, probably because it reports age, gender and ethnicity data, relies on individuals reporting their earnings and hours worked. "Administrative data," payroll data reported by employers, is more robust because it reports actual hours worked over shorter, more immediate periods and thus does not rely on individuals' potentially faulty or intentionally inaccurate reporting; rounding problems or gaps in the data that introduce measureable bias as the CBO itself describes at length in a different but similar paper (CBO 2008). This is mentioned but left unaddressed in the addenda to the 2011 Appendix a (20). CBO 2011 claims gaps in data arose through individuals' reporting for others not present (17) but that these gaps were filled by just assigning values imputed from similar individuals. CBO itself finds "matched administrative earnings avoids the apparent bias introduced by the imputed records" (2008) in a usefully similar study, and the effect was small but not trivial. These confounds are real, if not dramatic weaknesses underlying this paper.

More important to me are the choices CBO made restricting the group of wages they measured. The study does a thorough job of explaining why measuring compensation is not the goal of this study, but rather wages, which also explains the focus on individual rather than household earnings with some defensibility. Compensation includes fringe benefits; bonus earnings; options and other assets that are usually far more difficult to value and thus are not the goal of this study. But they did impute wage rates from non-wage workers in ways that are far from robust.

For the forty percent of workers not paid hourly wages, the survey data asked for weekly earnings and then divided that by hours worked to derive an imputed pay rate. These earnings were then weighted by number of hours worked, before they were sorted and ranked into percentiles such that "a worker's wage was represented in the distribution of wages in proportion to the number of hours he or she worked" (17). Since we have already excluded compensation and are considering wages, this seems reasonable up to the definition of 'hours worked.' Where hours worked were unknown, that week's earnings were divided by the prior week's hours. This may not be an accurate treatment of weekly earnings for any employee in an emergency decision making position like a business manager, public official or intellectual worker who is never really off the clock, and experiences longitudinal or intangible returns from hourly work.

Without getting into performance pay which could probably be ruled out as 'compensation,' a public official quoted in the Sunday paper would be 'at work' on a nominal day off and for the permanent future; an academic who achieves an insight while grocery shopping, develops it on the drive home and implements it in the classroom the next day; we can construct infinite examples, but CBO 2011 fails to address this vulnerability which probably confounds a significant portion of the 40% of responses that weren't paid by the hour. If these are higher-earning positions, they are even more represented in the study than they would be had hours at work (the denominator) been, more accurately, 'all the time.' This complements a rising trend toward unpaid work taken home, some 10.19 million hours in 2004, 55% of which were performed to "finish or catch up on work," with another 32% of unpaid home work explained by "nature of the job" (BLS 2004). This is not the unpaid work in family businesses already ruled out of 2011 paper's data or the self-employment excluded because survey takers couldn't gauge the share of capital investment to wages in microbusiness ownership (17). The report indicates capital gains, which are excluded from this analysis in general as neither wage or compensation earnings, are in fact increasing as a share of aggregate national income (4).

CBO 2008 finds the importance of non-wage self-employment earnings compared to wage earnings as not significant enough to change the results of the study that paper presents and points out that incorporated self-employment earnings will be reported as wages and therefore are probably included in the data the 2011 study describes. Hipple (2010) finds the absolute level of unincorporated self-employment largely stable if shifting toward wage-counted incorporated self-employment, but also reports the scale of this sector as comprising just under 11% of total national earnings from work.

The exclusion of all these types of earnings supports inquiry into the validity of data built on potential composition problems, the weighting for part and full time earnings. While the median is the proper measure of central tendency in cases of non-normal data; outliers etc., and it is meaningful to say that the distance between the top and the middle increased more than the distance between the bottom and the middle for different reasons, which is in a global sense the outcome of this research, the composition of earnings levels could change very drastically in ways policymakers may or may not want, while the median remains unchanged. As with mandatory performance of unpaid work at home, composition factors compromise the validity of describing a category of earnings the significance of which is falling in a changing workplace.

Other factors compound this compounding problem, particularly using age as a "proxy for experience" (17). Mosisa and Hipple point out that the labor force participation rate for individuals 55 and over, i.e. those between 55 and 64 in the sample this study analyzes, has been markedly increasing since 1995 (Mosisa 2006). If seniors switch from full-time work at the top of their earnings potential to lower-earning part-time work after 'retirement' (demonstrated in CBO 2008 for the 50-55 cohort), this completely undermines the point of using age as a proxy for experience, particularly when the largest cohort of the highest-earning but least-educated sector of the workforce (2011 p. 6) but yet that transition to part-time work is minimized in the weights against the remaining hourly wage earners. Likewise if a worker had two (or more!) part-time jobs, those earnings would not be represented as heavily as full-time earnings. My objective is not to solve this problem here but to point out that any conclusions drawn from this research deserve close critical scrutiny as far as the applicability of any assertions. Any claims supported by this research may very well have a falling, rather than increasing utility in a real policy application.

While these problems may not destroy the validity of the conclusions this report asserts, which is not my goal here, they force us to accept the possibility that there may be significant other factors at play. While the conclusions from this study may accurately describe wage hours, the earning patterns of the group who work the most hours at the highest pay rates, if that group is dwindling in importance relative to the society as a whole, describing them without a wider benchmark describes individual trees while leaving the wider forest unmapped. The increasing incidence of part-time work, self-employment, capital gains earnings (4) and gray/black market employment (CBO 2008) all combine to squeeze the utility of describing this class of earnings into an increasingly narrow channel. The other factors are growing in importance rather than shrinking, relative to this group of workers. This all leads to a broader possible criticism of the relevance of these conclusions even if the data are valid for the sector they describe.

To conclude this digression on the validity of the data, leaving discussion of relevance until after other meaningful areas for improvement have been mentioned, I will simply point out that the data in this report are not as useful as they could be because the complexity of all these adjustments restrict, rather than promote inquiry. The data this study presents are already weighted and sorted into percentiles, and while we can use them to check the validity of say claims about global trade effects they report (10; see Table 1, Appendix I) or their choice of business cycle peaks as indicating wage levels (13; see Table 2, Appendix I), reconstituting the raw data to compare them with the weighting adjustments will be so difficult that the noneconomist will very likely not be able to assess the conclusions thereby drawn and discussed. CBO could have done a better job assessing alternative interpretations, like they did in CBO 2008. There is more evidence in support but this is enough to raise valid questions.

Stepping outside of the data set to consider the relationship of these conclusions to their context, i.e. The relevance of the conclusions this report asserts (that the weighted median earnings have moved away from the lowest and highest percentiles they consider, for different reasons and in different ways), does not require us to search for factors outside CBO's own analysis. The factors they compare these conclusions against include other earnings, treated at length within the data set above; market and institutional effects which each have several components; education; gender; technology and inflation effects that happen in the broader context in which the data they discuss are embedded. These discussions achieve varying levels of success, and some degree of obfuscation.

The separation of market and institutional affects into discrete categories and assigning affects from these forces into separate buckets is an unnecessary simplification that clouds rather than clarifies, and ends up simply inaccurate. The report spends extensive space exploring market-based factors like skill-biased technological employment demand changes over time; labor force participation rates and educational attainment for women relative to men; and effects on U.S. workforce demographics from immigration and global trade, to a degree of detail that stands to its credit. But claiming that these factors occur outside the institutions that define the parameters under which all market transactions occur, grievously misrepresents Congress' own role in limiting the factors they allege drive these natural equilibria. Firewalling minimum wage and unionization effects into a realm of pure institutional change ignores effects across the entire marketplace the paper elsewhere asserts are the results of these institutional forces (Note 31, p. 15). This direct self-contradiction is the result of asserting that trade, labor, educational and microeconomic decisions about wage choice and marginal substitution between acceptable / available employment options etc. occur in some abstract vacuum unaffected by policies enacted by the authors of the paper themselves, is the result of this reductio ad absurdum and unfortunately misrepresents important conclusions to a readership which may lack the sophistication to recognize this division fallacy, namely those setting the policies determining those market factors, i.e. Congress. We hear for example that "growth in the supply of college graduates" offset demand pressures following World War II (6), but the fact that much of that supply arose because of the G.I. Bill (a policy within which market forces became nested) is left ignored.

These simplifications run throughout and distort the results this paper describes. On the macroeconomic level, while CBO 2011 discusses effects of worker immigration and the effects of these workers' educational endowments (12), there is no mention whatsoever of effects from illegal immigration on the distribution and level of earnings in the data set. Discussion of this type of under-the-table 'gray market' employment is glaringly absent, particularly from a discussion of institutional and market drivers on employment and wages. CBO 2008 cites evidence the scale of this type of earned income may be up to 2% of total national earnings and concludes both administrative and survey data miss this type of income. The 2011 report finds the effects of these immigration and global trade trends on employment negligible, after creating an example where low-educated workers actually stimulate employment in other trades (12). Perhaps we can impute illegal immigration effects from this example, but whether these are marketplace or institutional effects remains muddled.

I discussed distortions introduced by choice of survey vs. administrative data within the data set above; those decisions also distort the relevance of that data to the context it takes place within. Choosing individual rather than household earnings also overlooks microeconomic decisions consumers make in the substitution both of work against leisure; different types of employment against each other; off- or on-balance sheet employment; and between substitutes not represented in chained inflation adjustment. The CBO mentions the context of these wage changes relative to productivity and non-wage earnings (2011, p. 4), but in a clinical, abstracted way that leaves room for improvement.

CBO 2008 describes at some length how discussing household income accounts for earnings outside the parameters of the 2011 focus on wage income. CBO 2011 dedicates a paragraph to discussing the relation of this data to overall productivity (4). I assert however along the lines of the 2008 study, that these broader, general yet microeconomic factors affect individual substitution between employment criteria. The 2011 paper discusses in Note 5 (2) how the recent recession forced many workers to take wage employment at lower rates than the jobs they lost against their preference. This type of choice affects the relative level of income compared to individual consumption choices which affect the real earnings rate in complex but important ways. While the PCEPI reflects substitution effects between goods when those goods' prices change (17) and this type of chaining allegedly accounts for inflation-induced substitution better than the fixed-price CPI which only updates prices every decade, neither method compare larger modal consumption shifts forced by earnings change, between imperfect substitutes.

Simplistic analogies often illustrate complex topics adequately enough for discussion. Say a wage earner lost desired work and was unable to afford car insurance, payments and fuel, and so switched to public transportation, such that the net effect on purchasing power ended up positive. Or say a consumer took a roommate rather than live alone under earnings-driven economic stress. These constructed but plausible stories demonstrate substitution effects not generally reflected in even Fisher-type inflation chaining (PCEPI; see Fixler and Jaditz for the equations underlying the difference between CPI and PCEPI.) but probably have significant cause and effect on employment choice; substitution of work for leisure; availability of market baskets within which to choose real substitutes; i.e. real market forces determining workers' wage demands and educational decisions that CBO 2011 claims fundamentally drive those wage levels and their growing dispersion. A missed credit card payment double the cost of borrowing, which in effect introduces inflation to the purchased goods that probably shows up in one person's prices but noone else's. Stories of asset-rich but cash-broke overinvested home flippers dominate the financial press and demonstrate crippling inflation effects probably not accounted for in chaining indexes. I am trying to establish plausibility, not demonstrate proof.

Likewise this paper excludes significant other non-wage income as outside its purview, which is perfectly appropriate in an endogenous perspective limited to this group, but the relevance or utility of that exercise (and the expenditure of resources it undoubtedly entailed) becomes debatable when considered in light of other factors ignored here, but been described as important by the authors in similar recent papers (above). So we've discovered once and for all that median wages are increasing relative to floor wages; that the ceiling is rising relative to both middle and floor; that these effects have different causes among and results on different worker demographics. Considered in isolation, and even generously granting the data perfect validity, the question I end up with becomes, "So what." Unfortunately the answer seems to be "that variability causes true economic hardship for some people" (CBO Director Orszag, 2008), described at length at the beginning of a recession these 2011 conclusions hopefully bookend. But of course we have to be careful not to derive or imply policy recommendations from such impartial, objective results, like persistent lower wages for women over the last 30 years (CBO 2011, p. 4).