DATA 103 eral districts and reflect regional commuting flows. During the observation period, there are 97

regional planning units in Germany, 75 of which are located in West Germany. I calculate a population weighted average of the local unemployment rates of all districts in the commuting area, excluding the own district, and define the resulting rate as spatial unemployment rate.

Alternatively, I calculate a spatial unemployment rate that is the population weighted average of all first neighbors, i.e. of all districts that have a common border with the district of interest.

This measure does not account for commuting flows and is used for the purpose of robustness checks.

(8.4,17.1]

(6.1,8.4]

(4.2,6.1]

[1.8,4.2]

Source: Federal Statistical Office, own calculation.

Figure 4.1: Local unemployment rates, West Germany 2008

4.4 Econometric Framework

Theoretical considerations

In the literature cited in chapter 4.2, one repeatedly finds that men’s wages respond stronger to changes in the local unemployment rate than women’s wages. This implies that the gender wage gap is also potentially related to the level of unemployment. In a situation where a wage gap in favor of men exists, an increase in the level of unemployment would stronger affect men’s wages and should therefore reduce the wage differential. Consequently, I expect a negative relationship between the local unemployment rate and the gender wage gap. My theoretical considerations are thereby based on two assumptions that I derive from the literature outlined in section 4.2.

A1 The local unemployment rate and the level of wages are negatively correlated (Blanchflower and Oswald, 1994)

A2 Men’s wages respond stronger to changes in the unemployment rate than women’s wages (Baltagi and Blien, 1998)

Based on the assumptions (A1) that the unemployment rate and the level of wages are negatively correlated, and (A2) that men’s wages are more elastic, one can rewrite the gender-specific wage curves (equations 4.1 and 4.2) as the difference in mean wages as follows

ln(w_ir^m) = X_ir^mβ^m+γ^mln(Ur) +^m_ir (4.1)

denotes the log wage of men (women). i is an individual subscript and r a regional one. X_i^m (X_i^f) are male (female) characteristics and ^m_i (^f_i) is the error term in the wage regression. γ^m (γ^f) describes the effect of unemployment on men’s (women’s) wages. Writing the gap as the differential between mean wages

ln(w^m/f_r )

as in

4.4. ECONOMETRIC FRAMEWORK 105 equation (4.3), the resulting negative coefficient, (γ^m−γ^f)<0, makes sense if we built on the assumptions of the efficiency wage theory: In times of high unemployment the wage premium can be reduced because fewer vacancies in the labor market go along with diminished outside options and may weaken the negotiation power of employees. If men’s wages are more elastic than women’s, employers may pay a lower wage premium for men, while for women’s wages the adjustment is less pronounced. Therefore the gender-specific wage differential narrows.

As shown in the wage curve literature, not only the local unemployment rate is theoretically supposed to affect the wage level, but also the spatial unemployment rate might influence wages.

This is particularly true if we look at small labor markets with strongly marked commuting structures to adjacent regions. Thus, I hypothesize that a relationship between the spatial unemployment rate and the gender wage differential exists both at the regional level as well as within firms. The direction of the effect is theoretically less clear and will be investigated empirically in this paper. Moreover, in case of the intra-firm gender wage gap, organizational characteristics should be influential as well. Especially variables related to the wage setting process such as the existence of a works council and collective bargaining could moderate the effect of the unemployment rate.

Calculation of the gender wage gap

To investigate the theoretical considerations, I define measures for the gender wage gap at the district and firm level, which are later used as dependent variables in regression models. The simplest measure for the gender wage gap is the raw gender wage gap, i.e. the observed gap between gender-specific mean wages. At the regional level this is given for the r = 1, . . . , R districts int= 1, . . . , T years by

raw gap_rt= ln(w_rt^m)−ln(w^f_rt). (4.4)

The intra-firm wage gap is calculated analogously for thej= 1, . . . , J firms in year t= 1, . . . , T as

raw gap_jrt = ln(w^m_jrt)−ln(w^f_jrt). (4.5)

Part of this observed gap may be due to differences in the employees’ human capital endowments across districts or firms. Using the method introduced by the seminal papers of Oaxaca (1973) and Blinder (1973), I decompose the raw gender gap in the mean wages into a part resulting from differences in observable characteristics (explained part) and a part that reflects different evaluations of these characteristics (unexplained part). The unexplained part can be interpreted as the share of the gap resulting from different wage functions. It reflects the gender wage gap that would result if women had the same observable characteristics as men, i.e. the human capital-adjusted gender wage gap. The decomposition can generally be written as

∆ˆ^O

At the district level, the adjusted wage gap results from

adj. gap_rt= ˆ∆^S_rt=

for the r = 1, . . . , R administrative districts (Kreise/Kreisfreie St¨adte). It is calculated sepa-rately for each yeartand results inR·T different wage gaps. The ˆβ^m_rt are estimated in a Mincer regression at the district level

ln(w_itr^m) =β₀+β_rt^mX_irt+_irt, forr= 1, . . . , Rand t= 1, . . . , T. (4.8)

As explanatory variables, I use education, experience (quadratic), job tenure, occupational group, firm size and sector.

The intra-firm gender wage gap is decomposed analogously for the j = 1, . . . , J firms in each yeart= 1, . . . , T. The underlying wage function is estimated in gender-specific wage regressions at the district level with additional firm fixed effectsλ_jrt

ln(w_ijrt^m ) =β0+β_rt^mXijrt+λ^m_jrt+ijrt, forr= 1, . . . , Rand t= 1, . . . , T. (4.9)

4.4. ECONOMETRIC FRAMEWORK 107