disk: Synthetic Distributional Controls in Stata

In the field of impact assessment of public policies or strategic business changes, the synthetic controls method has established itself as an essential tool. Traditionally, this technique allows an artificial control unit to be constructed from a weighted combination of untreated units, thus estimating the causal effect on an aggregate variable, such as GDP, sales, or employment rate. However, the classic approach focuses only on the mean, leaving out valuable information about how the treatment affects the entire distribution of the variable of interest. This is where the concept of distributional synthetic controls comes in, recently implemented in Stata through the disk command. This innovation, based on the work of Gunsilius (2023), makes it possible to construct complete synthetic distributions for the treated unit, offering a much richer and more nuanced view of the causal effects.

The metaphor of the disc is illustrative: while a traditional synthetic control delivers a single value (such as a point in space), the distributional version displays a full spectrum of probabilities, similar to the way a compact disc stores multiple tracks. In practice, the researcher has data at the aggregate level but also at a finer granularity, such as household income, production per plant or individual responses in a survey. The disk command takes the empirical distributions of the control units and combines them by means of optimal weightings —obtained through a process of minimization of distance between the pre-treatment distributions— to generate a synthetic distribution that replicates the counterfactual behavior of the treated unit. The treatment effect is then defined as the divergence between the observed and synthetic distribution, either through differences between quantiles or by comparing cumulative distribution functions.

This approach is particularly valuable when the impact of a change is not homogeneous. For example, a tax reform can increase inequality even if the average income remains stable; an innovation incentive program can disproportionately benefit larger companies; or an adjustment in the pricing policy may alter the dispersion of demand. With the distributional method, analysts can break down the effect at different points in the distribution, identifying winners and losers. In addition, the disk package offers robust inference procedures using bootstrap and permutations, and provides clear visualizations that make it easy to communicate results to non-technical audiences.

From a business perspective, these capabilities open up immense opportunities for data-driven decision-making. Organizations looking to assess the impact of an organizational change, marketing campaign, or new technology adoption can benefit from distributional analysis that reveals differential effects across customer segments, regions, or teams. However, implementing this methodology in a corporate environment requires not only statistical knowledge, but also adequate technological infrastructure to handle often large and heterogeneous volumes of data. This is where collaborating with an expert software development partner is key.

At Q2BSTUDIO, we understand that advanced analytics comes to life when integrated into robust, scalable systems. Our specialty in custom applications allows us to build platforms that automate the extraction, transformation, and loading of data from multiple sources, directly feeding algorithms from synthetic distributional controls. In addition, we offer artificial intelligence for companies that powers these models, adding layers of prediction and recommendation. For example, a retail customer could use Disco to assess the impact of a promotional campaign across the entire distribution of spend per customer, and then implement an AI-based agent-based intelligent assistant that adjusts bids in real-time based on identified segments.

The integration of AWS and Azure cloud services is another fundamental pillar. The bootstrapping and permutation processes required by disk can be compute-intensive, but when deployed in the cloud they become efficient and elastic. With our expertise in cloud infrastructure, teams can run hundreds of iterations without worrying about local capacity, and store the results in secure data lakes. Similarly, cybersecurity is critical when handling sensitive customer or employee data; We implement pentesting and encryption protocols to ensure that information remains protected throughout the analysis cycle.

Another relevant dimension is the visualization of results. The disk command generates charts by default, but for executive reports or interactive dashboards, an additional layer of business intelligence is often needed. Our Power BI services allow you to integrate the outputs of synthetic distributions into dynamic dashboards, where decision-makers can explore quantile effects, filter by periods, or compare scenarios. In addition, when the analysis needs to be integrated into recurring workflows, we choose to develop custom software that automates everything from data collection to the update of synthetic models, ensuring that the company always has up-to-date evaluations.

A case of frequent use in the public sphere is the evaluation of subsidy or job training programs. With disk, a government could determine not only whether the program increased the average income of beneficiaries, but whether it reduced inequality within the treated group. In the private sector, an insurance company could analyze how a new pricing policy affects the distribution of claims, identifying whether risks are concentrated in certain profiles. These analyses, when combined with business intelligence services, become real competitive advantages.

It is important to note that the implementation of the distributional method is not trivial: it requires a careful selection of control units, verification of the validity of the distributional parallelism assumption, and choice of appropriate distance metrics. The disk command simplifies much of the process, but interpreting and communicating the results still requires statistical expertise. For this reason, at Q2BSTUDIO we not only provide the technology, but also support in the definition of the analytical strategy and in the validation of the models.

In short, the arrival of the album to Stata represents a qualitative leap in causal evaluation. By moving from averages to distributions, researchers and analysts gain a deeper understanding of the effects of interventions. For businesses, this capability translates into better-informed decisions, more precise segmentation, and evidence-based management. And when technology must be robust, scalable and secure, having an ally that offers everything from AWS and Azure cloud services to custom application development, including artificial intelligence and cybersecurity, makes the difference between an isolated project and a permanent analytical capacity. At Q2BSTUDIO we are prepared to accompany that journey, transforming cutting-edge statistical methods into tangible business solutions.