Filling vacancies: an empirical analysis of the cost and benefit of search in the labour market

Abstract

This paper investigates empirically the factors that affect the cost–benefit ratio of employers’ search. The empirical analysis is based on a small Dutch data set containing individual information on filled vacancies. It is found that firms that use advertisements during recruitment are sensitive to labour market conditions; their search cost per applicant rises (drops) in tight (slack) labour markets because of the diminished ability of advertisements to generate applicants in tight labour markets. Furthermore, it is found that the high search cost incurred by posting identical vacancies is more than compensated for by the benefits from having a larger flow of applicants.

Acknowledgements

We are grateful to Intomart b.v. in Hilversum for kindly providing the data set.

Notes

Job seekers face the same difficulties in locating vacant positions.

Rees (1966) classifies recruitment methods into two broad classes: formal and informal. Formal recruitment methods include the public employment service, the use of advertisement, the use of private employment agencies (headhunters), and the use of temporary help agencies. Informal recruitment methods include the use of open applications, the use of friends and relatives, internal recruitment, employee referrals, and referrals from other business contacts. The use of social networks to convey vacancy-related information generally characterizes informal recruitment methods.

Firms – by using contemporaneously more recruitment methods – can increase the gains from search by gathering a larger and more diversified pool of applicants (Kirnan et al., 1989).

Van Ours and Ridder (1993) find that advertisements significantly increase the rate of arrival of applicants and that they are quite effective in matching employed job seekers. Conversely, the public employment service is costless but personnel managers often complain that applicants lack the necessary motivation. Moreover, Barron and Mellow (1982) find that applicants from the public employment office tend to turn down employment offers and thus the search costs per vacancy increase. Informal recruitment methods are considered very effective because the employers and applicants can exchange reliable information about each other (Montgomery, 1991; Simon and Warner, 1992; Granovetter, 1995; Leicht and Marx, 1997). On the other hand, personnel managers often complain that informal recruitment methods generate a too low number of applicants (Mencken and Winfeld, 1998).

Recruitment method use affects vacancy duration as well; Roper (1988) finds that the use of multiple recruitment methods shortens the vacancy duration; the use of advertisements lengthens the vacancy duration, while the use of informal recruitment methods leads again to short vacancy duration.

Of course, the quality of applicants also matters.

The literature on the hiring activities of firms has mainly focused on the duration of such procedures (Beaumont, 1978; Van Ours, 1989; Roper, 1988; Burdett and Cunningham, 1998). Other studies have focused on the screening of applicants (Van Ours and Ridder, 1992, 1993; Abbring and Van Ours, 1994), and on the (sequential) use of recruitment methods (Russo et al., 1997; Gorter and Van Ommeren, 1999).

This was the case in the Netherlands towards the end of the 1990s when many firms complained about the scarcity of applicants (especially in the ICT, and care and education sectors). Similarly, Kölling (2002) reports that 47% of German employers ascribed difficulties met during the recruitment process to the lack of suitable candidates.

This would be more so if one could account for the opportunity cost of an open vacancy due to foregone profits. Such a cost component would be, clearly, procyclical.

Here we follow a consolidated tradition in empirical labour economics that includes the recruitment methods among the regressors, see Roper (1988), Van Ours and Ridder (1993) and Holzer (1996).

The Dutch unemployment rate fell monotonously from 1995 until 2001 when it began to rise again.

Notice that the stream of benefits deriving from search depends on the number of applicants and on their quality (productivity); therefore, ideally, one would like to produce a quality adjusted measure of the search cost per applicants (to generate very productive applicants may be expensive). To do so, however, one would need at least some information on applicants’ characterisitics that, in the data set at hand, are completely absent.

In a stationary environment where the rate of arrival of applicants arrive according to a known probability distribution the ratio of TC to TN equals the search cost per expected applicant, which is constant over time. Because of the stationary search environment, the marginal costs per applicant are equal to average costs per applicant, which in turn equate the marginal benefits per arrived applicant. In a stationary environment, the ratio TC/TN does not depend on the duration of the recruitment process because the passage of time increases the numerator and the denominator proportionally. Moreover, if firms with multiple vacancies search sequentially, the search costs and the arrival rate of applicants are both proportional to the number of open vacancies. Therefore, in a stationary environment with sequential search, the search costs per applicant should not depend on the number of open vacancies.

The vacancy duration can capture foregone profits only in the case of an open vacancy. If the recruitment process takes place entirely during the advance notice period the foregone profit element does not apply. However, we cannot distinguish between recruitment actions that took place, entirely, during the advance notice period and those that extended beyond that period. At any rate, the vacancy duration should correlate with the screening and selection costs.

The use of OLS in a duration analysis is appropriate when, as in our case, the data are a flow sample and do not suffer from censoring, Lancaster (1992).

In addition, in a non-stationary search environment in which the condition for optimal search is set such that the marginal costs equal the marginal benefits may still hold per period (for example, per week) if the structural parameters are constant during that period. So, the reservation productivity and the amount of search effort are set at the beginning of each period, but they may change from period to period (e.g. week to week). This implies that the average search costs per applicant (TC/TN) are a weighted average (over time) of the changing (marginal) search costs per (expected applicant); therefore, they become a function of completed vacancy duration.

Ideally, one should estimate a system of three equations, one for the search costs, one for the number of applicants and the third for the vacancy duration. The identification of the system of equations should preferably be achieved through the usual exclusion restriction. However, in our case we were not able to find any defendable exclusion restriction. Alternatively, a reduced form SURE approach could have been adopted but this would have been equivalent to the estimation of three separate equations because all three equations use the same set of regressors.

The Dutch Central Bureau of Statistics puts the growth rate of the GDP deflator in the years 1994–1995 until 1998–1999 to 2.0%, 1.2%, 2.0%, 1.7%, 1.6%, respectively.

1 Dutch Gulden (1 NLG) is worth 0.45 Euros.

These results might change once the quality of the applicants generated could be accounted for.

We have also introduced interaction terms between the recruitment methods and the yearly dummy in all the three models estimated. Advertisements were the only recruitment method that showed a significant interaction effect.

We have also experimented with a quadratic specification for the number of identical vacancies and for the number of vacancies filled in the previous year in all three models. The quadratic specification could always be rejected against the linear one.