Models and measurement in economics
Models and measurement in economics. Do macro-economic modelers and macro-economic statisticians talk about the same variables?
[This post benefitted from comments by Josh Mason and Diane Coyle.]
Variables like ‘capital’, ‘the price level’, ‘unemployment’ and ‘consumption’ are routinely mentioned in macro-economic models as well as measured by macro-statisticians. But are the economic modelers and the economic statisticians, while using the same words, also using the same definitions or even talking about comparable concepts? Not always. And differences are sometimes fundamental (Cristiano, 2011, especially paragraph 2.1). In a series of posts of which this is the introductory one I will argue that:
* many persistent, non-trivial differences exist between the concepts and definitions of the variables as used in the models and the concepts and variables as defined and measured by the statisticians, especially between the modern ‘DSGE’ models (more about these below) and the macro-statistics, i.e. the statistics compiled within the framework of the ´broad´ national accounts;
* these differences matter for theoretical as well as practical purposes.
DSGE models are, in their basic form, non-monetary ´Robinson Crusoe´ models which model the entire economy as a single person or, to be more precise, as one single homo economicus. More sophisticated versions exist but the concepts of the variables are pretty much based upon this basic form. The ´broad national accounts´ are fundamentally monetary models which estimate different kinds of monetary flows (wages, profits, consumption, investments) between different sectors of the economy and which, contrary to the DSGE models, are based upon aggregation of individual micro transactions and which also enable the estimation of physical flows of products and hours and which increasingly also contain data on stocks of financial variables. In a subsequent post I will describe the models in more detail, here I will only stress that the differences are also reflected in the culture of economics. Models are often developed in universities by academics who earn a personal reputation based on this work and who are pressed to earn such a reputation – or to lose their job. The statistics are developed and produced in specialized institutions which often publish the results anonymously. There is limited interaction between these two worlds when it comes to either developing models or developing macro statistics. This occurs even to the extent that it might be necessary to clarify in this post, which might mainly be read by academic economists, that with ‘statistics’ I do not mean econometrics or Anova and Ancova tables but data on the price level, (un)employment and many other macro-economic variables which are routinely and in a remarkably consistent way estimated by statistical offices like Insee in France, the ONS in the UK, the Bureau of Labor Statistics (BLS) in the USA, the Ministry of Statistics and Programme Implementation in India, and, when it comes to monetary macro statistics, central banks (about this also Bos 2003 and Bos 2013).
At this point some of the readers might wonder why in economics non-trivial differences between model-variables and statistical variables exist at all.1 Isn’t a science supposed to be a consistent and coherent system of models and measurements instead of a hodgepodge of mutually inconsistent models and statistical variables?! Such a thought might idealize science a little but in macro-economics the difference is in my opinion so large that the ‘why?!’ question requires some attention. In the rest of this introductory post I will therefore highlight an episode in the history of (un)employment statistics and the role of unemployment in economic models to show that existing differences are neither coincidental nor haphazard but related to the persistent existence of different economic schools or paradigms. This also means that changing the concept of either the model variables or the estimated variables requires a ´paradigm change´. This is a task which, according to the philosophy of science, is not easily accomplished. The Keynes-Lucas discrepancy about involuntary unemployment mentioned in the next paragraph does not falsify that idea. But the episode also highlights that key decisions about the actual definition, operationalization and measurement of unemployment were not instigated by academic economists but by the, at that time, most powerful legislature of the world, the US congress. Never mind the differences between the paradigms, the politicians could not wait for economists to end their disputes and wanted to know about the ins and outs of measurement (which, as it happened, were to a considerable extent developed by economists from yet another inclination, the ´institutionalists´)! Gathering macro-economic statistics is costly and many of the variables measured (unemployment!) are politically sensitive, and it is not a coincidence that the activity has to be funded by the government. This makes the differences between the economists even more remarkable.
- The concept of unemployment. Models, measurement and politics.
According to Bos (2006, 2013) a (not necessarily chronological) sequence of:
- thinking about concepts;
- translating concepts into definitions;
- operationalization of the definitions; and
- measurement of the operationalized variables.
In this many stakeholders are involved as is characteristic for the historical development of measurements of (systems of) economic variables. If Bos is right while, at the same time, a difference between the meaning of variables in theoretical models and in economic statistics exists, some kind of disconnect between theoretical developments and the development of economic statistics must be traceable. This is the case. The development and definition of the concept of ‘involuntary unemployment’ might serve as a case in point. In 1936 John Maynard Keynes defined, in chapter II of The general theory of employment, interest and money his new concept of ‘involuntary unemployment’:
“Men are involuntarily unemployed if, in the event of a small rise in the price of wage-goods relatively to the money-wage, both the aggregate supply of labour willing to work for the current money-wage and the aggregate demand for it at that wage would be greater than the existing volume of employment.” (Keynes 1936)
According to Keynes, this situation could exist but it was inconsistent with the labour market theory of the ‘classical’ economists of those days.2
Keynes had good reasons to introduce and define this concept as in his time the classical ideas did not seem to work. Classical economists stated that in a market economy lower nominal wages would lead to more demand for labour and, hence, lower unemployment. This meant that (high) unemployment was either voluntary (when people did not accept lower wages) or ‘institutional’, i.e. caused by institutions and frictions which prevented a decline of wages. But in the UK of the twenties things were different. ‘Average earnings´ declined from a level of 1.03 in 1920 to 0.71 in 1923 and declined further from 0.73 in 1928 to 0.68 in 1932 (Dimsdale, Hills and Thomas 2010; Bank of England Internet 1). Despite these declines, average UK unemployment in the twenties was about as high as peak unemployment before 1914 while, after 1929, unemployment rose to totally unprecedented levels. High unemployment in the twenties could with a lot of tweaking and twisting maybe be explained by the argument that wages had not fallen enough. But the strong rise of unemployment after 1929 was, considering the declines which already had taken place, totally anomalous. Economists had something to explain. Keynes tried to do this – which led him outside of the classical framework. Or in fact: it led him to erect a new, larger framework where a ‘classical’ economy was only one of many possibilities and involuntary unemployment, an impossibility in the special case of the classical economic world, could exist.
For several decades, Keynes’ ideas took hold. In 1976 however the ‘New Classical’ economist Robert Lucas, widely regarded as the most influential macro-economist of the 1970-2008 period and one of the intellectual fathers of the DSGE models, explicitly responded to the Keynes challenge (Lucas 1976). He starts citing a 1933 Hayek quote which implies that in those days classical ‘general equilibrium’ economics indeed had something to explain (and was not able to do this). The quote is followed by an empirical description of business cycles which is based upon the ideas and measurements developed by a proud student of Thorstein Veblen, the economist Wesley Mitchell (and, subsequently, by Geoffrey Moore and Julius Shiskin (see Frumkin, 1998)): what is there to explain? So far, so good. But remarkably, Lucas leaves out all labour market variables which are part of this elaborate system of interrelated lagging, coincident and leading business cycle indicators. The labour market is simply not included in Lucas’s description of the business cycle and unemployment, involuntary or not, hence does not require explanation. This left Lucas of course with the problem of how to explain the clear and cyclical changes in measured unemployment. At this time, and in contrast to the days of Keynes, dependable and detailed statistics on (un)employment, showing the clearest of cyclical patterns, were published and widely discussed on a regular basis. Lucas did this by negating (not the same thing as disproving) the idea of ‘involuntary unemployment’, totally misrepresenting the contents of chapter 2 of the General Theory and equating ‘unemployment’ with ‘leisure’ not only as a matter of speech but also by explicitly using data on vacation and weekend behaviour of employed people to describe the behaviour of the unemployed. He goes on by stating that ‘measured unemployment’ is a clear choice by a ´worker/producer´ (compare a cobbler in an eighteenth century village) which in fact means that he disregards the institution of wage labour (note the word ´close´ in the next quote):
“If “leisure” is highly substitutable over time, he will work longer on high price days and close early on low price days… there is little evidence that much time is spent in job search … that measured unemployment measures any activity at all… . Indeed, I suspect that the unwillingness to speak of workers in recession as enjoying “leisure” is more a testimony to the force of Keynes’ insistence that unemployment is “involuntary” than a response to observed phenomena”.3
Keynes did of course not state that all unemployment is, by his definition, ´involuntary´. He stated that in certain situations unintended consequences at the macro level could thwart the micro efforts of the unemployed to get a job, which could give rise to involuntary unemployment. But Lucas’s ideas carried the day when it came to modeling unemployment in the DSGE models. As Lawrence Christiano stated 25 years later, in 2011 when commenting on a DSGE model which tried to estimate unemployment:
“First, I am skeptical that the people designated as unemployed in the model satisfy the official United States definition of unemployment. Second, the model implies that the unemployed are happier than the employed” (Christiano 2011).
Christiano shows this in a detailed way: what’s called unemployment in such models is the same as thatmeasured by economic statisticians and is at odds with our knowledge about the mental and material suffering of the unemployed. The implicit definition of ‘unemployment’ in the models happened despite a mayor mistake by Lucas: the suggestion that ‘measured unemployment’ does not measure any activity at all. The contrary is true. Activity is the very basis of the unemployment statistics. Which brings us from ‘high’ theory to the mundane world of statistical definitions, operationalization and measurement. According to Eurostat, which follows the International Labour Organization (ILO) guidelines:
“Unemployed persons are persons aged 15-74 who were without work during the reference week, but who are currently available for work and were either actively seeking work in the past four weeks or had already found a job to start within the next three months” (emphasis added, Eurostat Internet 1, see also Christiano 2011).
And this is not a new definition. When Lucas wrote his article, a comparable definition was already in use for quite some time, at least in the USA, with the explicit approval of the US congress. In 1962, the US government issued a report titled ‘Measuring employment and unemployment’. Congressional hearings about this report were held in 1963 (n.a. 1963). The introductory text of the hearings by Robert A. Gordon (the economist, not the sociologist) is remarkable:
“The actual timing of the Committee’s appointment was, almost certainly, influenced by the publication of an article by James Daniel which appeared in the September 1961 issue of the Reader’s Digest. The article was called “Let’s Look at Those ‘Alarming’ Unemployment Figures.” … an egregious example of irresponsible journalism. In effect, it charged that the official data on unemployment were being deliberately manipulated in order to justify larger Government spending and more extensive Government controls. While this article probably precipitated the decision to set up a committee of outside experts at that particular time, a much more basic set of forces had been at work for a number of years that would almost certainly have led the Federal Government eventually to seek a new appraisal of our labor force statistics”.
The unemployment statistics clearly were so important to the Congress that a hearing was held not about unemployment but about a question as arcane (and fundamental!) as the way (un)employment was measured! The testimonies give witness to the broadly perceived necessity to base macro unemployment data on individual situations which individual people actively are trying to change, to show the (at the micro level) involuntary nature of unemployment. This is what happened: a 1967 Bureau of Labor Statistics Report by Robert Stein shows that these remarks were taken to heart and incorporated in the survey questions which are used to measure ‘unemployment’ (Stein, 1967). Later the International Labour Organization (ILO) adopted comparable recommendations about the measurement of (un)employment, which via the ILO were disseminated to statistical offices all over the world (this process still goes on). This makes ‘measured unemployment’ as an ´observed phenomenon´ at least at the micro level ‘active’, ‘involuntary’ as well as a disequilibrium situation. Lucas was wrong about all of this. Despite this, ´new classical´ economists chose to negate the work of economic statisticians, the work on business cycles and in fact even the US Congress (Goldberg and Moye 1985 shows that especially during times of crises political interest in macro labor statistics is invariably high, the hearings were part of a pattern). The new (un)employment statistics also yielded information on ´broad unemployment´ (people without employment not actively seeking, involuntary part time employment etc.), which showed that the level of unused capacity was even larger than earlier perceived. Economists had even more to explain and they had more data than ever to do this.4 But new classical economists like Lucas chose to shy away from this task. About this one can also consult De Vroey 2004, a book titled Involuntary Unemployment which investigates this concept at great length but tellingly does not spend a title or iota on the concept and definition of measured unemployment…. An unfortunate consequence of this closing of the classical mind has been that, at this moment, no agreement or discussion exists about a measurable definition of ´involuntary unemployment´ (De Vroey 2004 dismisses the whole concept), which can be considered to be a mayor failure of institutional and Keynesian economists as well as of the economic statisticians.
The Stein article also contains some, from a methodological perspective, interesting discussions about the operationalization of the concept of unemployment, such as whether we should include 14 to 16 year olds or about the precise phrasing of questions. Without going into details: in both cases such seemingly arcane questions do matter for the outcomes of the measurement; ‘measured unemployment’ is not just dependent on the definition of the variable, but also on the way it is operationalized and on the method of measurement (sample size!). A final point: many economists were involved in the development of economic statistics in general and also in the development of (un)employment statistics. The preponderance of ‘institutional economists’ among the ‘fathers’ (and a remarkable amount of mothers) of economic statistics, many of them proud and self-conscious students or admirers of Veblen, the most radical critic of classical economics of his generation, is remarkable.5 The prime example is Wesley Mitchell but one can also mention Morris Copeland, the father of the flow of funds, and for labour statistics Isador Lublin, student and friend of Veblen and long-time head of the BLS (Ayres 1963; much more extensive Rutherford 2011 and Goldberg and Moye 1985). Academic economists (Keynesian and classical alike) seem to have left the actual measurement of macro-economic data to a remarkable extent to the ´second generation´ of ´institutional economists´6, and nowadays to the economic statisticians.
- The work ahead
I hope that, at this point, the reader will be aware of the existence of a serious disconnect between the conceptualization, definition, operationalization and measurement of macro-variables on one hand and with many economic models on the other. In the next post I hope to say a little more about ´material and methods´ as well as to provide a more in depth comparison of the two main models mentioned, after which I will proceed by trying to make a precise comparison between the concepts of the classical and new classical ´DSGE´ variables and the variables which are actually measured.
Ayres, Clarence (1963). ‘The legacy of Thorstein Veblen’ in: Dorfman, Joseph (ed.) Institutional economics. Veblen, Commons and Mitchell reconsidered pp. 63-94. University of California Press, Berkeley/Los Angeles.
Bank of England, Internet 1 Accessed 1 April 2016
Bos, Frits (2003). The national accounts as a tool for analysis and policy: past, present and future. Eagle Statistics, Berkel en Rodenrijs.
Bos, F. (2013), ´Meaning and measurement of national account statistics´. Paper provided at the Political Economy of Economic Metrics conference, available here.
Dimsdale, Nicholas, Sally Hills and Ryland Thomas (2010). ´The UK recession in context. What do three centuries of data tell us´. Bank of England Quarterly Bulletin 2010/Q4 pp. 277-291 available here
European Commission (March 2014), Quarterly report on the Euro Area 13 no 1. Brussels, available here.
Eurostat, Internet 1. Assed 4 April 2016.
Frumkin, Norman (1998), Tracking America’s economy. Armonk, New York.
Goldberg, Joseph and William Moye (eds.)(1985). The first hundred years of the Bureau of Labour statistics. U.S. government printing office, Washington. Available here
Keynes, John Maynard (1936), The general theory of employment, interest and money. Macmillan University press, Cambridge. Available here
Lawrence, Christiano (2011), ‘Comments on Gali, Smets and Wouters “Unemployment in an estimated New Keynesian model”’, available here
Lucas, Robert, ´Understanding business cycles´. Paper prepared for the Kiel Conference on Growth without Inflation, June 22-23 available here
n.a. (1963). ‘Measuring employment and unemployment. Hearings before the subcommittee on economic statistics of the joint economic committee of the congress of the US 88-th congress. June 6 and 7 1963’. U.S. government printing office, Washington, available here.
Mayhew, Anne (2010), ”Copeland on money as electricity”, real-world economics review 53, pp. 52-55, Available here
Rutherford, M. (2011), The institutional movement in American economics 1918-1947. Science and social control. Cambridge University press, New York.
Stein, Robert (1967). ‘New definitions for employment and unemployment’ in: Bureau of Labor Statistics, Monthly report on the Labor Force February 1967 pp. 1-25 available here
Vroey, Michel de (2004), Involuntary unemployment. Routledge, New York.
1 This situation seems to be improving. Internships of students of macro-economics at statistical institutes are however still not an obligatory or even advised part of their education.
2 Looking at (un)employment, domestic demand inflation and wage developments in Spain and Ireland and the estimates of the ´NAWRU´ rate of unemployment (Non Accelerating Wage Rate of Unemployment, which in these countries might after 2008 better be called the DWRU) it turns out that the Keynes definitions describe the situation in these two countries pretty well. See European Commission 2014, graph II 1.1.
3 The derogatory tone of the Lucas remarks seems to fit in a tradition. The remarks of Martin Gainsborough of the National Industrial Conference Board in the transcripts of the 1963 USA Congress hearings on ´Measuring employment and unemployment´ (n.a. 1963) contain comparable though less extreme statements.
4 This discussion is too USA centered to my liking. However the availability of material and the importance of USA economists in the ‘New classical’ revolution in macro-economics as well as the fact that in the case of (un)employment statistics USA economic statisticians seem to have led the way as well as ease of expose led me to leave it this way.
5 And of neoclassical economics – Veblen invented the very phrase!
6 Keynes himself was deeply immersed in the conceptualizing of the national accounts and had a kind of ´Veblen´ role in the UK.
From: pp.3-7 of World Economics Association Newsletter 6(2), April 2016