Unemployment Spells and Unemployment Experience By GEORGE A. AKERLOF AND BRIAN G. M. MAIN* According to a modem view of the labor market, as reflected most prominently in papers by Martin Feldstein, Robert Hall, and George Perry, among others, unemploy- ment in the United States is on average of quite short duration and, as a result, should be modeled not in terms of stocks of per- sons who are unemployed for fairly long periods of time, but rather in terms of flows of persons whose spells of unemployment are quite short. An important statistic which supports this view of unemployment is the large number of completed spells of unemployment in any one year. For example, in the year 1969, which is in no way atypical in this respect, 93,640,000 persons' were in the labor force some fraction of the time, and there were a reported 32,100,000 spells of unemployment with an average duration of only 4.6 weeks.^ This paper examines ways in which such statistics on the large number of spells and the correspondingly short durations are mis- leading in terms of the welfare implications of unemployment. Note that the average duration statistic just cited concerned completed spells of un- employment. It is not the same as the usu- ally quoted official government statistic, the average duration of spells currently in pro- gress (or interrupted spells). The difference between these two statistics is exactly analo- gous to the difference between mean life • Universit y of California-Berkeley. We would like to thank James Beny, Norman Bowers, Harvey Hamel, Steve Salant, and Janet Yellen for valuable comments and assistance. We also gratefully acknowledge the generous financial support of the Board of Governors of the Federal Reserve System and the U.S. Depart- ment of Labor under Small-Grant Research Project No. 91-06-78-27, administered by the Institute of Busi- ness and Economic Research of the University of Cali- fornia-Berkeley. 'See Table C-5, U.S. Bureau of Statistics (1970). ^See Kaitz, Table 2, p. 12. span (which is the average length of a com- pleted life) and the mean age of the popula- tion (which is the average length of lives currently in progress). This analogy between mean life span and mean age suggests that the average duration of spells in progress should underestimate the average length of completed spells because the duration of each spell in progress is less than its com- pleted length. However, there is also a bias in the other direction. In any sample of the population, a spell in progress has a proba- bility of being sampled which is propor- tional to its length, and thus the sampled average length of spells in progress may overstate the length of completed spells since the longer spells are proportionately over- sampled.'' The paper by Hyman Kaitz, which for the first time estimated the average length of completed spells for the United States, has been quite influential in supporting the new view of unemployment. His paper shows that the average length of a completed spell from 1948 to 1969 was only 40-62 percent of the average length of spells in progress. Table 1 brings Kaitz' estimates on the aver- age lengths of completed spells up to date using the method proposed by Steven Salant.^ It also compares the estimated aver- age completed spell lengths for these years to the official statistics on average in- progress (or interrupted) spell lengths. This paper concerns the interpretation of a spell of unemployment, and offers new information on the length of spells of unem- ployment separately for people who experi- 'For a good discussion of these issues, see Salant. ^According to Salant, if the density of interrupted spell lengths Tisg(T) = (r-l)a'-\a+T)-', the den- sity of completed spell lengths S is fiS) = ra''(a+ ^j-C'+i), Data on interrupted spells were used for maximum likelihood estimates of a and r. The function /(5) with these estimated values yields estimated densi- ties of completed spell lengths. 885