Inside IES Research

A key challenge when conducting cost analysis as part of an efficacy study is producing information that can be useful for addressing questions related to replicability or scale. When the study is a follow up conducted many years after the implementation, the need to collect data retrospectively introduces additional complexities. As part of a recent follow-up efficacy study, Maya Escueta and Tyler Watts of Teachers College, Columbia University worked with the IES-funded Cost Analysis in Practice (CAP) project team to plan a cost analysis that would meet these challenges. This guest blog describes their process and lessons learned and provides resources for other researchers.

What was the intervention for which you estimated costs retrospectively?

We estimated the costs of a pre-kindergarten intervention, the Chicago School Readiness Project (CSRP), which was implemented in nine Head Start Centers in Chicago, Illinois for two cohorts of students in 2004-5 and 2005-6. CSRP was an early childhood intervention that targeted child self-regulation by attempting to overhaul teacher approaches to behavioral management. The intervention placed licensed mental health clinicians in classrooms, and these clinicians worked closely with teachers to reduce stress and improve the classroom climate. CSRP showed signs of initial efficacy on measures of preschool behavioral and cognitive outcomes, but more recent results from the follow-up study showed mainly null effects for the participants in late adolescence.

The IES research centers require a cost study for efficacy projects, so we faced the distinct challenge of conducting a cost analysis for an intervention nearly 20 years after it was implemented. Our goal was to render the cost estimates useful for education decision-makers today to help them consider whether to replicate or scale such an intervention in their own context.

What did you learn during this process?

When enumerating costs and considering how to implement an intervention in another context or at scale, we learned four distinct lessons.

1. Consider how best to scope the analysis to render the findings both credible and relevant given data limitations.

In our case, because we were conducting the analysis 20 years after the intervention was originally implemented, the limited availability of reliable data—a common challenge in retrospective cost analysis—posed two challenges. We had to consider the data we could reasonably obtain and what that would mean for the type of analysis we could credibly conduct. First, because no comprehensive cost analysis was conducted at the time of the intervention’s original implementation (to our knowledge), we could not accurately collect costs on the counterfactual condition. Second, we also lacked reliable measures of key outcomes over time, such as grade retention or special education placement that would be required for calculating a complete cost-benefit analysis. This meant we were limited in both the costs and the effects we could reliably estimate. Due to these data limitations, we could only credibly conduct a cost analysis, rather than a cost-effectiveness analysis or cost-benefit analysis, which generally produce more useful evidence to aid in decisions about replication or scale.

Because of this limitation, and to provide useful information for decision-makers who are considering implementing similar interventions in their current contexts, we decided to develop a likely present-day implementation scenario informed by the historical information we collected from the original implementation. We’ll expand on how we did this and the decisions we made in the following lessons.

2. Consider how to choose prices to improve comparability and to account for availability of ingredients at scale.

We used national average prices for all ingredients in this cost analysis to make the results more comparable to other cost analyses of similar interventions that also use national average prices. This involved some careful thought about how to price ingredients that were unique to the time or context of the original implementation, specific to the intervention, or in low supply. For example, when identifying prices for personnel, we either used current prices (national average salaries plus fringe benefits) for personnel with equivalent professional experience, or we inflation-adjusted the original consulting fees charged by personnel in highly specialized roles. This approach assumes that personnel who are qualified to serve in specialized roles are available on a wider scale, which may not always be the case.

In the original implementation of CSRP, spaces were rented for teacher behavior management workshops, stress reduction workshops, and initial training of the mental health clinicians. For our cost analysis, we assumed that using available school facilities were more likely and tenable when implementing CSRP at large scale. Instead of using rental prices, we valued the physical space needed to implement CSRP by using amortized construction costs of school facilities (for example, cafeteria/gym/classroom). We obtained these from the CAP Project’s Cost of Facilities Calculator.

3. Consider how to account for ingredients that may not be possible to scale.

Some resources are simply not available in similar quality at large scale. For example, the Principal Investigator (PI) for the original evaluation oversaw the implementation of the intervention, was highly invested in the fidelity of implementation, was willing to dedicate significant time, and created a culture that was supportive of the pre-K instructors to encourage buy-in for the intervention. In such cases, it is worth considering what her equivalent role would be in a non-research setting and how scalable this scenario would be. A potential proxy for the PI in this case may be a school principal or leader, but how much time could this person reasonably dedicate, and how similar would their skillset be?  

4. Consider how implementation might work in institutional contexts required for scale.

Institutional settings might necessarily change when taking an intervention to scale. In larger-scale settings, there may be other ways of implementing the intervention that might change the quantities of personnel and other resources required. For example, a pre-K intervention such as CSRP at larger scale may need to be implemented in various types of pre-K sites, such as public schools or community-based centers in addition to Head Start centers. In such cases, the student/teacher ratio may vary across different institutional contexts, which has implications for the per-student cost. If delivered in a manner where the student/ teacher ratio is higher than in the original implementation, the intervention may be less costly, but may also be less impactful. This highlights the importance of the institutional setting in which implementation is occurring, and how this might affect the use and costs of resources.

How can other researchers get assistance in conducting a cost analysis?

In conducting this analysis, we found the following CAP Project tools to be especially helpful (found on the CAP Resources page and the CAP Project homepage):

• The Cost of Facilities Calculator: A tool that helps estimate the cost of physical spaces (facilities).
• Cost Analysis Templates: Semi-automated Excel templates that support cost analysis calculations.
• CAP Project Help Desk: Real-time guidance from a member of the CAP Project team. You will receive help in troubleshooting challenging issues with experts who can share specific resources. Submit a help desk request by visiting this page.

Maya Escueta is a Postdoctoral Associate in the Center for Child and Family Policy at Duke University where she researches the effects of poverty alleviation policies and parenting interventions on the early childhood home environment.

Tyler Watts is an Assistant Professor in the Department of Human Development at Teachers College, Columbia University. His research focuses on the connections between early childhood education and long-term outcomes.

For questions about the CSRP project, please contact the NCER program officer, Corinne.Alfeld@ed.gov. For questions about the CAP project, contact Allen.Ruby@ed.gov.

This is the second of a two-part blog series from an IES-funded partnership project. The first part described how the process of cost-effectiveness analysis (CEA) provided useful information that led to changes in practice for a school nurse program and restorative practices at Jefferson County Public Schools (JCPS) in Louisville, KY. In this guest blog, the team discusses how the process of conducting value-added analysis provided useful program information over and above the information they obtained via CEA or academic return on investment (AROI).

Since we know you loved the last one, it’s time for another fun thought experiment! Imagine that you have just spent more than a year gathering, cleaning, assembling, and analyzing a dataset of school investments for what you hope will be an innovative approach to program evaluation. Now imagine the only thing your results tell you is that your proposed new application of value-added analysis (VAA) is not well-suited for these particular data. What would you do? Well, sit back and enjoy another round of schadenfreude at our expense. Once again, our team of practitioners from JCPS and researchers from Teachers College, Columbia University and American University found itself in a very unenviable position.

We had initially planned to use the rigorous VAA (and CEA) to evaluate the validity of a practical measure of academic return on investment for improving school budget decisions on existing school- and district-level investments. Although the three methods—VAA, CEA, and AROI—vary in rigor and address slightly different research questions, we expected that their results would be both complementary and comparable for informing decisions to reinvest, discontinue, expand/contract, or make other implementation changes to an investment. To that end, we set out to test our hypothesis by comparing results from each method across a broad spectrum of investments. Fortunately, as with CEA, the process of conducting VAA provided additional, useful program information that we would not have otherwise obtained via CEA or AROI. This unexpected information, combined with what we’d learned about implementation from our CEAs, led to even more changes in practice at JCPS.

Data Collection for VAA Unearthed Inadequate Record-keeping, Mission Drift, and More

Our AROI approach uses existing student and budget data from JCPS’s online Investment Tracking System (ITS) to compute comparative metrics for informing budget decisions. Budget request proposals submitted by JCPS administrators through ITS include information on target populations, goals, measures, and the budget cycle (1-5 years) needed to achieve the goals. For VAA, we needed similar, but more precise, data to estimate the relative effects of specific interventions on student outcomes, which required us to contact schools and district departments to gather the necessary information. Our colleagues provided us with sufficient data to conduct VAA. However, during this process, we discovered instances of missing or inadequate participant rosters; mission drift in how requested funds were actually spent; and mismatches between goals, activities, and budget cycles. We suspect that JCPS is not alone in this challenge, so we hope that what follows might be helpful to other districts facing similar scenarios.

More Changes in Practice 

The lessons learned during the school nursing and restorative practice CEAs discussed in the first blog, and the data gaps identified through the VAA process, informed two key developments at JCPS. First, we formalized our existing end-of-cycle investment review process by including summary cards for each end-of-cycle investment item (each program or personnel position in which district funds were invested) indicating where insufficient data (for example, incomplete budget requests or unavailable participation rosters) precluded AROI calculations. We asked specific questions about missing data to elicit additional information and to encourage more diligent documentation in future budget requests. 

Second, we created the Investment Tracking System 2.0 (ITS 2.0), which now requires budget requesters to complete a basic logic model. The resources (inputs) and outcomes in the logic model are auto-populated from information entered earlier in the request process, but requesters must manually enter activities and progress monitoring (outputs). Our goal is to encourage and facilitate development of an explicit theory of change at the outset and continuous evidence-based adjustments throughout the implementation. Mandatory entry fields now prevent requesters from submitting incomplete budget requests. The new system was immediately put into action to track all school-level Elementary and Secondary School Emergency Relief (ESSER)-related budget requests.

Process and Partnership, Redux

Although we agree with the IES Director’s insistence that partnerships between researchers and practitioners should be a means to (eventually) improving student outcomes, our experience shows that change happens slowly in a large district. Yet, we have seen substantial changes as a direct result of our partnership. Perhaps the most important change is the drastic increase in the number of programs, investments, and other initiatives that will be evaluable as a result of formalizing the end-of-cycle review process and creating ITS 2.0. We firmly believe these changes could not have happened apart from our partnership and the freedom our funding afforded us to experiment with new approaches to addressing the challenges we face.   

Stephen M. Leach is a Program Analysis Coordinator at JCPS and PhD Candidate in Educational Psychology Measurement and Evaluation at the University of Louisville.

Dr. Robert Shand is an Assistant Professor at American University.

Dr. Bo Yan is a Research and Evaluation Specialist at JCPS.

Dr. Fiona Hollands is a Senior Researcher at Teachers College, Columbia University.

If you have any questions, please contact Corinne Alfeld (Corinne.Alfeld@ed.gov), IES-NCER Grant Program Officer.

IES would like to congratulate and thank David Card, Joshua D. Angrist, and Guido W. Imbens, who received this year’s Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel. The work of these laureates has greatly contributed to the ability of researchers to provide causal evidence in support of education practice and policy decision making. IES is proud to have previously supported Card and Angrist in some of their education research work.

Many key issues in education cannot be analyzed using randomized experiments for practical and ethical reasons. Card’s work (with Alan Krueger) on natural experiments helped open up a novel approach to providing causal findings. In natural experiments, outcomes are compared for people who have differential access to a program or policy (or a change in a program or policy) because of real life conditions (for example, institutional or geographic differences) rather than through random assignment by researchers. Natural experiments have been adopted by IES grantees to examine a broad variety of education programs and policies such as PreK expansion, early literacy, school choice, school turnaround programs, high school curriculum change, and changes to postsecondary remediation course requirements. Angrist and Imbens showed how to estimate a causal treatment effect when individuals can choose to participate in a program or policy, which often occurs in natural experiments and can occur in randomized experiments after researchers have randomly assigned participants. IES grantees widely use their instrumental variable approach for both experimental (often involving designs based on school lotteries) and quasi-experimental designs.

In addition to developing evaluation designs and methods that have been broadly applied within education research, Card and Angrist have also directly carried out education research important to the field, sometimes with the support of IES. For example, Card is a principal investigator (PI) on two IES-funded studies on gifted education (elementary school and middle school) and is a co-PI on the National Center for Research on Gifted Education. Angrist is PI on two IES-funded studies, one on charter schools and one evaluating a Massachusetts desegregation program.

Angrist and Imbens have also supported the work of IES. Both researchers served as IES peer reviewers on grants and reports, and Imbens provided the What Works Clearinghouse with advice on standards for regression discontinuity designs (RDD) and co-authored one IES-supported paper regarding RDD (a method that has also become widely used in IES-funded research).

IES thanks Card, Angrist, and Imbens—both for their contributions to causal methods and for their direct participation in education research—and congratulates them for this recognition.

Education research is often a slow and costly process. An even more difficult challenge is replicating research findings in a timely and cost-effective way to ensure that they are meaningful for the wide range of contexts and populations that make up our nation’s school system.

In a recent op-ed, IES Director Mark Schneider and Schmidt Futures Senior Director for Technology and Society Kumar Garg pitched the idea that digital learning platforms may be a way to accelerate the research enterprise. These platforms will enable researchers to try new ideas and replicate interventions quickly across many sites and with a wide range of student populations. They could also open the door for educators to get more involved in the research process. For example, Learn Platform supports districts as they make decisions about purchasing and implementing products in their schools, and ASSISTments provides infrastructure for researchers to conduct cheaper and faster studies than they would be able to do on their own.

IES Director Mark Schneider and NCER Commissioner Liz Albro recently attended a meeting sponsored by Schmidt Futures focused on these issues. Two major takeaways from the meeting: first, there is already progress on building and using platforms for testing interventions, and, second, platform developers are enthusiastic about integrating research capabilities into their work.

As we consider how we can support platform developers, researchers, and education personnel to co-design tools to enable more efficient, large scale research on digital learning platforms, several questions have arisen:  

1. What digital learning platforms already have a large enough user base to support large scale research studies?
2. Are there content areas or grade levels that are not well supported through digital learning platforms?
3. What are the key features that a platform needs to have to support rigorous tests and rapid replication of research findings? 
4. What are the barriers and challenges for companies interested in participating in this effort?
5. What kinds of research questions can best be answered in this research environment?
6. What kind of infrastructure needs to be developed around the platform to enable seamless collaborations between education stakeholders, researchers, and product developers?

We know there are some of you have already given these questions some thought. In addition, there are other questions and issues that we haven’t considered. We welcome your thoughts. Feel free to email us at Erin.Higgins@ed.gov and Elizabeth.Albro@ed.gov. And join NCER’s Virtual Learning Lab in their virtual workshop “Designing Online Learning Platforms to Enable Research” on April 17th, 3:00pm-5:00pm Eastern Time. Learn more about the workshop here.

Data sharing agreements are critical to conducting research in education. They allow researchers to access data collected by state or local education agencies to examine trends, determine the effectiveness of interventions, and support agencies in their efforts to use research-based evidence in decision-making.

Yet the process for obtaining data sharing agreements with state or local agencies can be challenging and often depends on the type of data involved, state and federal laws and regulations regarding data privacy, and specific agency policies. Some agencies have a research application process and review timeline available on their websites. Others may have a more informal process for establishing such agreements. In all instances, these agreements determine how a researcher can access, use, and analyze education agency data.

What are some guiding principles for successfully obtaining data sharing agreements? 

Over several years of managing projects that require data sharing agreements, I have learned a few key principles for success. While they may seem obvious, I have witnessed data sharing agreements fall apart because one or more of these principles were not met: