The determinants of urban (un)employment duration: evidence from Barcelona

Fu et al. (1993) and Dendir (2006) analyse the determinants of unemployment duration in urban areas of developing countries. None of them address the effects of the spatial mismatch.

There are a variety of studies on the duration of Spanish unemployment (see e.g. Ahn and Ugidos-Olazabal 1995; Bover et al. 2002; Jenkins and García-Serrano 2004; Alba-Ramírez et al. 2007 and references therein). In contrast, studies on the duration of both employment and unemployment seem to be limited to (García-Fontes and Hopenhayn 1996; García-Perez 1997; García-Perez and Muñoz-Bullón 2005).

For example, since distant workers are more likely to have short employment durations, one may wonder whether these jobs are genuinely temporary (thus supporting the mismatch hypothesis) or just stepping stones to long-term contracts (thus questioning the mismatch hypothesis). In this respect, Johnshon (2006: 355) reports evidence suggesting that blacks and Hispanics in MSAs “are more willing to accept a job and continue searching, given the lower offer rates for unemployed blacks and Hispanics relative to unemployed whites”.

Along the same lines, Wheeler (2001: 880) interprets the negative estimates reported by Alperovich (1993) on the relation between city size and the length of unemployment spells (long-term job seekers out of total population) in a cross section of non-Arab Israelis cities as evidence that “cities facilitate search”.

However, some studies examining urban unemployment durations do not address this issue. For example, Fu et al. (1993) estimate the distribution of unemployment spells in Shanghai using a Gaussian kernel estimator and Dendir (2006) estimates a lognormal accelerated failure time model using data from households in seven major urban centres in Ethiopia. In our case, we cannot address the effects of the spatial mismatch due to the lack of appropriate data (see Sect. 4). The lack of job accessibility measures is in fact the main difference between our model specification (essentially based on previous Spanish national studies reviewed in the next subsection) and that typically used in previous urban unemployment duration studies reviewed here (who, with the exception of Rogers (1997), do not use a logistic discrete hazard as we do; notice, however, that she does so in a competing risks framework). In this respect, it seems important to stress that although this paper does not seek to analyse the effects of the spatial mismatch, results reported by Matas et al. (2010) suggest that its effects in the city of Barcelona during our period of analysis are likely to be very weak.

See also McGregor (1977) for an early analysis using log-linear models and UK data from the city of Pasley.

Yet another set of studies use (waves of) survey data, either from the Labour Force Survey (the EPA, e.g. Bover et al. 2002) or the Survey of Conditions of Life and Work (the ECTV, e.g. Ahn and Ugidos-Olazabal 1995). “However the ECVT, like the EPA, did not have information about benefit levels. Moreover, it was a cross-sectional survey and unemployment spell data were collected by retrospective recall questions” (Jenkins and García-Serrano et al. 2004: 240). On the other hand, these surveys included household information (ECTV) and covered all workers (EPA). Lastly, notice that none of these sources enable the effects of the mismatch hypothesis on the duration of (un)employment to be tested; see, in contrast, results in Matas et al. (2010) on the probability of females being employed using census data (individual characteristics being the main covariates considered).

An important exception is the work of (Matas et al. 2010), who find evidence of spatial mismatch in the metropolitan area of Barcelona (and Madrid). Notice, however, that their probit estimates were obtained using 2001 census data and a wide definition of the metropolitan area. Moreover, reported statistics show that poor job accessibility and residential segregation are particularly severe in recent years (e.g. the percentage of jobs in the central city was ten percentage points larger in 1981) and in the suburban areas of the metropolitan area (i.e. those far from the city centre). Therefore, the effects of the spatial mismatch in the city of Barcelona should be substantially lower in the 1980s and early 1990s.

We found a few cases of benefits related to temporary disability. Rather than distinguish them as a different category, we decided to include them as assistance benefits.

See Sect. 2. See also the Appendix for details on the definitions and data sources we use.

However, individual labour factors differ in the employment and unemployment specifications because they refer to the previous spell, i.e. the previous unemployment/employment spell duration and the previous/current unemployment benefits are determinants of the current employment/unemployment spell. Also, in the employment model, we set to zero both the unemployment spell duration and the unemployment benefits prior to the first contract we observe. Since we have the work history of the individuals, this seems a plausible imputation.

There were no exits from employment at \(T=1\), so in the employment model, we either constructed the dummies from \(T=2\) or did not include the dummy for \(T=1\). We found that the first option produced better results in terms of AIC values. We also explored a constant duration for the first two or three periods of employment and a piecewise specification based on the months intervals reported in Table 3. However, these approaches resulted in much worse AIC values.

These results are available upon request. Coefficients and standard errors were also not stable when using a “complementary log-log” model (see e.g. Cameron and Trivedi 2005) with either gamma-distributed or nonparametric frailty.

Interestingly, this is consistent with the matching process being more efficient in urban labour markets (Wheeler 2005, Finney and Kohlhase 2008).

We speculate that the higher rate of activity, the sectorial specialisation and the use of temporary contracts (see Sect. 3.2) may lie behind these differences. However, the flexibility and demand of skilled workers that characterise urban labour markets are other factors worth considering (Zenou 2000; Glaeser and Maré 2001).

The reference groups match the ones used by García-Perez (1997). Thus, hazard estimates of employment correspond to a lower-middle educated young man in the fourth quarter of a year affected by the 1984 reform in the following macroeconomic environment: 2.56 % GAV growth rate, 17.58 % unemployment rate and 0.59 % unemployment growth rate. Also, hazard estimates of unemployment correspond to a lower-middle educated young man without insurance benefits in the third quarter of a year affected by the 1984 reform in the following macroeconomic environment: 2.28 % GAV growth rate, 18.5 % unemployment rate and 0.88 % unemployment growth rate. To facilitate comparisons, the scale used for probabilities (in the employment hazard) and durations is the one used by García-Perez (1997).