The Analysis of Symmetrically Predicted Endogenous Subgroups (ASPES) Method

The Analysis of Symmetrically Predicted Endogenous Subgroups (ASPES) Method is a tool for understanding the impact of an intervention on “endogenous subgroups” using experimental evaluation data. Endogenous subgroups are those that are defined by events that occur after the point of random assignment. For example, program managers and evaluators may be interested in whether the impact of an intervention is larger for the subgroup of study participants who experienced more of the intervention (and possibly smaller for those who experienced less of it) (Peck, 2013). In this example, the ASPES method can be used to understand these “dosage effects” by estimating the impact of the intervention on treatment group members who received a high dosage of the intervention relative to their counterparts in the control group that would have received a high dosage of the intervention had they been offered the treatment. Beyond this example, the ASPES method can address a wide array of questions related to the impact of an intervention on study participants who have varying post-random assignment experiences.

The purpose of this website is to provide instruction on two versions of the ASPES method: (1) a discrete version that allows researchers to estimate impacts on discretely-defined subgroups and (2) a continuous version that allows researchers to estimate the relationship between a continuously-defined endogenous variable and impact magnitude.

The Empirical Challenge Addressed by the ASPES Method

When estimating impacts on subgroups defined by an observed baseline characteristic it is common practice to simply divide the sample according to some binary or multichotomous variable (e.g., gender or age), then to estimate impacts separately for each subgroup. Similarly, researchers may be interested in understanding how program impacts vary for sample members who experience various paths after random assignment. However, estimating the impact of an intervention on the subgroup of treatment group members who received a high dosage and control group members who would have received a high dosage poses an empirical challenge, because treatment dosage is not defined (and therefore not observed) for control group members.

Given an experimental setting, where study participants are randomly assigned to treatment and control conditions, the expectation is that any subset of one experimental group has a counterpart in the other. For example, treatment group members who receive a high dosage of the intervention will have counterparts in the control group who would have received a high dosage of the intervention had they been offered the treatment. On average, this high dosage subgroup (which includes both high dosage treatment group members and would-be high dosage control group members) will have the same measurable and unmeasurable characteristics in both the treatment and control groups, as would any experimentally defined subgroup (Moulton, Peck, & Greeney, 2017). Based on this insight, the ASPES method established in Peck (2003) uses baseline characteristics to construct subgroups with high propensities for a post-randomization choice, event, experience, or path.

What research questions can the method address?

The discrete version of the ASPES method can address the following general question: What is the effect of an intervention on discretely-defined endogenous subgroups?

The continuous version of the ASPES method can address the following general question: What is the effect of a continuously-defined endogenous variable on intervention impacts?

Peck (2013) details the following classes of questions that can be addressed by the ASPES method:

  • Potential effects on “no-shows”
  • Effects by treatment dosage or quality
  • Effects of individual components of a multi-faceted intervention
  • Effects on subsets of the control group that make particular fall-back choices when denied access to the intervention

What are the data requirements?

The ASPES method requires data from an experimental evaluation that includes the following measures:

  • a treatment group indicator
  • an outcome of interest
  • a measure of the endogenous variable of interest
  • and baseline data for predicting the endogenous variable

What resources are available on this website?

This website contains the following resources for researchers interested in learning more about the ASPES method:

  • A description of the discrete and continuous versions of the ASPES method.
  • Materials from a Society for Research on Educational Effectiveness (SREE) workshop on the ASPES Method. This workshop provides instruction on the ASPES method and provides practical examples. Slides and a recording of the presentation are included.
  • The SPI-Path|Individual User-Guide, which provides detailed practical guidance and SAS and Stata code for those interested in implementing the discrete version of the ASPES method.
  • SAS and Stata code templates for researchers interested in applying the continuous version of the ASPES method.
  • A comparison of ASPES to alternative methods used for mediation analysis (instrumental variables and structural equation modeling) in terms of the research questions addressed, estimation process, interpretation of estimates, and key assumptions.
  • A list of references on the ASPES method and its applications.