Using Prediction Gaps to Understand the Unequal Influence of Social Contexts on Educational Outcomes

Add to calendar
to

The first three weeks (1-20 July) of the ODISSEI Incubator for Lifecourse Modelling will be hosted at the CCSS. Every Friday there will be a presentation that is open to the general public and will take place at the CCSS living room (MIN4.16) from 12:30-13:30.

Using Prediction Gaps to Understand the Unequal Influence of Social Contexts on Educational Outcomes
Dr. Javier Garcia-Bernardo
4th of July 12:30-13:30
Social contexts—such as families, schools, and neighborhoods—shape individuals’ outcomes. The key question is not simply whether they matter, but rather for whom and under what conditions they matter more. Here, we argue that prediction gaps—differences in predictive performance between statistical models of varying complexity—offer a lens for identifying surprising empirical patterns (i.e., not captured by simpler models) that should be reconciled with existing sociological theories on the effects of social contexts. Using population-scale administrative data from the Netherlands, we compare logistic regression, gradient boosting, and graph neural networks to predict university completion from early-life contexts. Overall, prediction gaps are small, suggesting that previously identified indicators, particularly parental socioeconomic status, capture most of the measurable variation in educational attainment. However, gaps are larger for disadvantaged students, indicating that the effects of social context for these groups go beyond simple models in line with sociological theory. Our paper show the potential of prediction methods to complement traditional sociological theory building.

The ODISSEI Incubator for Lifecourse Modelling
For the month of July, ODISSEI will be conducting an incubator where a multidisciplinary group of researchers will work in collaborative groups to establish new standards and data models for analyzing lifecourse data in the Secure Environment at Statistics Netherlands. Lifecourse research has huge potential for understanding a wide range of complex social outcomes from fertility through to political polarization. However, we have seen in other fields that developing such models requires intensive coordination across a research field, the setting of benchmarks, and the establishment of common standards. This incubator aims to encourage this level of coordination and standardization whilst fully respecting and enforcing the stringent and necessary security protocols that are in place when working with this data. By the end of the incubator it is hoped that the teams will have agreed on, developed, and deployed common and effective approaches for modelling lifecourse processes that are usable not only in the context of the Netherlands but in equivalent data environments. To supplement the work of the incubator and broaden engagement, participants and external speakers will provide presentations on their work every Friday at lunchtime. These talks are open to any interested parties who might like to learn more about the work of the incubator. The results of the Incubator will be presented at the ODISSEI Conference in November.

Other presentations from this workshop that are open to the general public:

 

Start date and time
End date and time
Location
Physical Meeting >> CCSS Living Room, Room 4.16, Minneartgebouw
Entrance fee
FREE
Add to calendar