Skip to contents

My working paper, Commercial Embeddedness and Sanctions Busting: Intra-Industry Trade, Dependence, and Economic Asymmetry, is available on APSA Preprints. It studies trade between third-party countries and countries under sanctions.

The paper asks how three features of a trading relationship fit together. Do the countries trade within the same industries? How dependent is the third party on trade with the sanctioned country? Are the two economies similar in size, or very different?

These are also useful questions for binwise. Categories can help us describe combinations that would be difficult to follow by looking at one variable at a time. We can then ask a separate question about how well models using those categories predict outcomes.

What the categories mean

  • Intra-industry trade (IIT). The paper separates zero IIT from positive IIT, then divides positive IIT into lower and higher groups. Zero means no measured intra-industry component, not no trade relationship.
  • Trade dependence. The paper groups the target’s share of the third party’s trade using the 50th, 75th, and 90th percentiles. This gives more detail among the most dependent relationships.
  • Economic size. The paper separates more similar economies, moderately different economies, and highly asymmetric economies using the difference in their logged GDP values.

The percentile boundaries describe this sample. They are not established points at which economies suddenly start behaving differently.

Why grouping helps show the pattern

Consider relationships in the higher positive-IIT group. Among similarly sized economies, the observed rate of sanctions busting rises sharply across the dependence categories. Among highly asymmetric economies, the rate is much lower in the two highest dependence categories.

The full case study shows the heatmaps, explains each square, and includes the observation counts. One square records a 70.7% rate among similar economies with higher positive IIT and the highest dependence. That means 3,469 of the 4,909 observations in that combination are coded as sanctions busting. It does not mean that this group accounts for 70.7% of all sanctions busting.

The benefit of grouping here is that the combinations are easier to describe and compare. A useful descriptive grouping does not automatically make the best prediction model.

What the additional prediction checks show

For binwise, we also compared models on observations left out of fitting. These checks were prepared for the package case study and should not be read as additional results reported in the working paper.

All three models predict the same outcome for the same test observations. Observations involving the same unordered country pair stay together when the data are divided into training and test groups.

Model Brier score
Original numbers with a linear interaction 0.0480
Categories 0.0420
Original numbers with flexible curves and interactions 0.0363

A Brier score measures the average squared difference between a predicted probability and the observed outcome, coded zero or one. Smaller values mean more accurate probability predictions by this measure.

The categories improve on the linear comparison here. The flexible model improves further. Its specification was written for this paper’s analysis; it is not identical to adding spline_df = 3 to a basic binwise call.

These results do not undo the descriptive value of the categories. They show why we should distinguish making a pattern easier to discuss from choosing the model that predicts most accurately. Neither comparison, by itself, establishes a causal effect.

Follow the analysis

Open the full paper case study for the category boundaries, heatmaps, confidence intervals, and prediction comparisons. That guide uses small summaries of the real data included with binwise. It recreates the figures from saved results; it does not rerun the original regressions.

You can also read the working paper for the research argument, definitions, and analysis.

Cite the working paper

Preble, Keith. 2026. Commercial Embeddedness and Sanctions Busting: Intra-Industry Trade, Dependence, and Economic Asymmetry. APSA Preprints. Posted November 16, 2026. https://doi.org/10.33774/apsa-2026-7wxvk.