Econometric model

three-categorical degree relatedness measure is further condensed into a binary variable, un-related. The outcome variable is equal to one if the current job is not adequate (or unrelated) to the field of study and equal to zero otherwise, that is, if a job is somewhat or entirely adequate (or related). Table 2.2 shows that the middle category does not differ much across either field of study or gender. Because of that we dichotomizing the variable (ratings of 5 and 4 as an unrelated job,yit = 1, and ratings of 1 - 3 as a related job, y_it = 0). This simplification eases the interpretation of the results. To show the robustness of the results, we will further show an estimation in which (i) the middle category is coded as an unrelated occupation, and (ii) the 5-likert scale variable is not simplified.

2.4 Econometric model

We aim to examine whether females with a STEM degree are less likely to enter the job market with a STEM occupation or not and whether this is most pronounced for females with children. Thus, we will examine whether female STEM graduates are not entering STEM occupations at a higher rate than men, relative to other professional fields. The baseline regression equation is as follows:

y_it=α₁+γ₁ f emale_i+δ₁ ST EM_i+ρ₁ ST EM_i f emale_i+β₁⁰ X_it+_it (2.1) where index i stands for the individual and t for the wave. The dependent variable y_it is a binary variable that is equal to one if an individuali holds a job unrelated to their degree in time t and equal to zero if the job is related.

On the right-hand side of Equation (2.1),female is a dummy variable equal to one if the observed individual is a woman and equal to zero if the individual is a man. The binary variableSTEM is equal to one if an individual has a degree in STEM and zero for indi-viduals holding degrees from all other fields of study. The classification of STEM degree

Chapter 2. 2.4. Econometric model subjects follows the classification of the Federal Employment Agency (Bundesagentur f¨ur Arbeit,2019).³ We include the interaction of the two dummy variables in the regres-sion equation to isolate the specific effect for female STEM graduates. The coefficient on ρ1 hence gives the field-specific gender difference. Adding δ1 and ρ1 shows the changed probability of working in an unrelated job specifically for female STEM graduates. A positive value of this sum would indicate that female STEM graduates not entering STEM occupations excessively. In contrast, the sum’s negative value would indicate a lowered entry rate of females compared to males within STEM.

The vector of variables X_it includes several covariates to control for other factors that might affect the attrition from the STEM field. The covariates are derived from both the literature on job mismatch and STEM entry behavior. Table 2.1 lists the full set of covariates. We include demographic information, (work) experience before graduation, socio-cultural variables, personal and educational background, as well as job character-istics, study information, and origin information. Additionally, following Hunt (2016), we include dummies for all other areas of studies.⁴

We further follow Hunt (2016) by separating STEM into the groups engineering and computer sciences (EngComp) and mathematics and natural sciences (MatNat). For this, we replace the STEM dummies with a dummy for EngComp and for MatNat. The regression equation looks as follows:

y_it =α₂+γ₂ f emale_i+δ₂ EngComp_i+ρ₂ (EngComp_i·f emale_i)

+µ₁ M atN at_i+µ₂ (M atN at_i·f emale_i) +β₂⁰ X_it+η_it (2.2) To investigate possible fertility effects, we further add a children-dummy into the equa-tion and interact the dummy with the variables for the EngComp degree and gender

3The Federal Employment Agency includes the fields mathematics, physics, chemistry, pharmacy, biology, geo-sciences geography, as well as, computer sciences, and all engineering fields in STEM.

4Results are robust to not adding all other areas of study.

Chapter 2. 2.4. Econometric model and their interaction:

y_it =α₃+γ₃ f emale_i+δ₃ EngComp_i+ρ₃ (EngComp_i·f emale_i) +λ₃ (EngComp_i·f emale_i·children) +γ₄ children +γ5 (childreni·f emale_i) +γ6 (EngComp_i·childreni)

+µ₃ M atN at_i+µ₄ (M atN at_i·f emale_i) +β₃⁰ X_it+ξ_it (2.3)

For the variable children, we use three different specifications. To evade possible en-dogeneity of children and occupation decisions, our main variable children is a binary variable equal to one if an individual has one or more children before graduation and zero otherwise. Concentrating on children born before leaving university mitigates the potential endogeneity because we know childbirth happened before obtaining the degree.

Otherwise, entering the job market and the willingness to get a child might be simul-taneous. Then, we further use the variablechildren^W which is additionally equal to one in the second wave when the graduate got children before that wave. Lastly, to check for robustness, we further neglect the timing of birth and just set the variable equal to one if the graduate has one or more children in the respective wave. This, however, has to be interpreted with great caution.

The interaction ofEngComp and female still isolates the field-specific gender difference that is specific for female EngComp graduates, but now only for childless women. The coefficient of the triple interaction,λ₃, shows if the field-specific gender difference effect for women with children. Thus, ifδ₃+ρ₃+λ₃ > δ₃+ρ₃ >0, women with an EngComp degree would over-proportionately not enter their occupational field because of childcare obligations, relative to men, relative to other professional fields and relative to those without children.

The job-entry and wish to have children might also be affected by the relationship-status of individuals. Employees might fear that young women will have children in the near future. Moreover, women might already be planning to have children soon

Chapter 2. 2.5. Empirical results