A LTHOUGH many early models of job search assumed that job seekers were unemployed, economists have come to recognize the importance of on-the-job search.' Mattila (1974), for example, found that roughly 60% of quitters experienced no unemployment between jobs, implying that the majority of voluntary job changers engaged in on-the-job search. The choice of search mode (i.e., employed vs. unemployed search) has been modelled in the theoretical search literature (Barron and McCafferty, 1977; Burdett, 1978); however, there have been no attempts to estimate empirically the determinants of this choice.2 Some researchers have, however, examined the resulting wages associated with these two types of search and found a greater wage return to employed than to unemployed search (Mattila, 1969; Black, 1980). However, because voluntarily unemployed search involves forgoing income (Burdett, 1978), while employed search does not (abstracting from voluntary reductions in hours), incomemaximizing job seekers would never quit to search unless the expected net return to unemployed search exceeded that for employed search. Thus, the previously mentioned empirical findings at first glance might appear to contradict the assumption of income maximization by job seekers, because some workers do quit into unemployment.3 In this paper, it is argued that previous attempts to compare the returns of employed and unemployed job search suffer from a selectivity problem. Specifically, searchers who accept offers without experiencing any unemployment are likely to have unobservable variables (such as informal job contacts and ability) which will raise the relative return to employed search compared to that of voluntarily unemployed job seekers,4 providing an explanation for higher wages for on-the-job searchers than for unemployed searchers. This paper estimates the relative wage effects of employed and unemployed search and the determinants of the choice of search mode (i.e., employed vs. unemployed search). Unlike earlier work (Mattila, 1969; Black, 1980), the previously mentioned selectivity problems are accounted for. This is the first attempt to estimate the determinants of choice of search mode.5 Without adjusting for selectivity bias (as regards employed vs. unemployed search), it is found that employed search is associated with greater expected wage offers than unemployed search, controlling for measurable personal characteristics. However, accounting for selectivity, the expected wage offer given unemployed search is greater than that given employed search. Finally, the greater is the difference between the expected offer given unemployed and that given employed search, the more likely one is to choose unemployed search. These findings suggest that quitting to search unemployed may be a wealth-maximizing decision. Received for publication November 5, 1980. Revision accepted for publication August 17, 1981. * University of Illinois at Urbana-Champaign and Arizona State University, respectively. Professor Low is partially supported by a Faculty Grant from the Graduate College at Arizona State University. The authors thank Francine Blau, Wallace Hendricks and two anonymous referees for comments and suggestions. I For a summary of this literature, see Lippman and McCall (1976). 2 Barron and McCafferty ( 1977) do estimate an aggregate time-series model of the log of the proportion of quits entering unemployment as a function of the vacancy rate (proxied by the help wanted advertising index). However, such a model ignores search that does not lead to quits and does not include variables beyond the vacancy rate that influence an individual's choice of search mode. I It is also possible that the forgone earnings due to unemployed search could in principle be less than the possible savings in direct search costs due to unemployed search (relative to employed search)-see Burdett (1978); in this case, a lower wage return to unemployed search could imply that quitting was based on income considerations. Further, individuals may have quit for nonwage reasons. 4 Black ( 1980) found that, other things equal, those who planned to seek new employment but did not quit had significantly lower wage growth than those who did not plan to search and did not quit. This finding strongly suggests the likelihood of a selectivity bias (Black, 1980, p. 227). ' Black's (1976, 1980) data (the Michigan Panel Study of Income Dynamics-PSID) for his work on employed search did not allow him to distinguish employed and unemployed search.