This replication package contains three sets of codes:
- STATA code replicating the full analysis that draws on proprietary data provided by Statistics Netherlands (STATNL).
- Simple STATA code that uses a synthetic dataset to illustrate the mechanics of our model, demonstrate its relation to the conventional FBOE model, and provide an example of statistical inference for composite coefficient estimates.
- Even simpler SPSS code that does the fundamentals of 2.
The first Stata code (FBOE_FFE_STATNL.do) documents complete data generation, model estimation, and result extraction workflows corresponding to our empirical analysis that draws on proprietary administrative data provided by STATNL. Proprietary nature of this data means that the analytical dataset is not supplied in this replication package. The only way to get access to the data is to secure the necessary funding and start a new research project at Statistics Netherlands. Inquiries regarding data access should be addressed to microdata@cbs.nl.
The code for the STATNL analysis was written and executed in STATA 16.0, OS Windows 10. Note that you will need Stata 16.0 and higher in order to execute this code, otherwise you will have to adjust the code by removing all the 'frame' commands (used for generating figures and output tables). All supplementary packages are provided with the code.
To execute the code with proprietary STATNL data, make sure that you have access to the following datasets:
GBAPERSOONTAB
GBAVERBINTENISPARTNERBUS
KINDOUDERTAB
To run the principal analysis, please execute the do-file (FBOE_FFE_STATNL.do). Before you do so, please change the global MAIN_FOL macro in the preamble of this do-file to your project folder (that is, the folder containing the do-file FBOE_FFE_STATNL.do and the subfolder 'auxiliary_scripts' with the supplementary packages). The preamble also lists further instructions.
The code is commented and it contains additional information that should facilitate the replication efforts.
The second Stata code (FBOE_FFE_synthetic_data.do) uses a synthetic dataset (which roughly mimics the properties of the original proprietary dataset) to estimate our preferred logistic regression model,
and demonstrate its relation to the conventional FBOE model and its coefficients,
The code also illustrates how to test significance of composite coefficient estimates mentioned in the manuscript. This is shown on an example of the coefficient estimate corresponding to the hypothetical scenario of having one fewer younger brother, and one more older brother (this coefficient is akin to an alternative definition of the FBOE recently used by Blanchard & Lippa, 2021).
The code was written and executed in STATA 16.0, OS Windows 10. Unlike the STATNL code, it will run on any version of Stata. Please note that the code automatically downloads the synthetic dataset from our online repository. There is no need to locate or download the dataset before running the code.
The SPSS script (FBOE_FFE_synthetic_data.sps) uses the same synthetic dataset to estimate several specifications of the model. Also in this case, the code automatically downloads the synthetic dataset from our online repository.
The script is constrained to bare essentials, because SPSS is a more restrictive software than other commonly-used statistical packages. The key limitation is that SPSS does not allow for testing of composite coefficients. The most tractable way to test these composites is to re-arrange the model specification so that the tested coefficient appears directly in the regression equation.
This is again illustrated on the example of the FBOE coefficient akin to Blanchard & Lippa (2021). To test its statistical significance, we have to estimate an adjusted version of our preferred model specification with numbers of brothers replaced by numbers of sisters:
The coefficient is what we're looking for. The ceteris paribus condition implies that this coefficient approximates the increase in the probability associated with increasing the number of older siblings by one, while keeping the sibship size and the number of older and younger sisters fixed, which is equivalent to having one fewer younger brother and one more older brother.
For reference, here is a list of correspondence equations for the three listed sets of coefficients: