There’s a concept that shows up everywhere in education, is generally understood at a surface level, and yet is still deeply underappreciated when it comes to policy, accountability, and public narratives.
Selection effects.
Elite institutions aren’t elite because they teach more effectively
Take higher education. Elite universities are considered elite largely because they reject most of their applicants. Their status comes from who they don’t admit far more than from anything special happening on campus.
This is widely understood within higher education. And yet, in public discourse, we often talk as if these institutions are producing better outcomes because they have discovered some superior instructional approach.
They haven’t. They are enrolling students who already have strong academic preparation, family support, financial stability, and social capital. Outcomes are explained by selection effects, not instruction.
The “best” K–12 schools are often just the best-situated ones
The same dynamic plays out in K–12.
Many public schools labeled as “the best” are located in high-SES communities. Sometimes that comes with higher funding, but far more often it comes with other advantages including:
- Stable housing and food security
- Low student mobility
- Access to tutoring and enrichment
- Parents with time and flexibility
None of this diminishes the work of educators in those schools. But it matters enormously when we compare outcomes.
When students show up consistently, supported, and ready to learn, schools look effective. When students are mobile, hungry, or dealing with instability outside of school, schools look less effective—even when teachers are doing extraordinary work.
This isn’t a controversial observation, but it still shows up inconsistently in accountability systems and related conversations.
A blended learning lesson I didn’t fully understand at first
Years ago, I led a project looking at “proof points” in blended learning. This wasn’t a large or long-term research effort. We were examining outcomes data from blended learning implementations inside traditional school districts, typically in brick-and-mortar schools. These implementations were usually limited to a subset of a school, such as a grade level, a subject area, or a small pilot.
We were looking for examples of success backed by data, and we found some. In fact, many of these implementations showed significantly positive academic outcomes in the first year. Over time, however, a pattern emerged that I didn’t fully understand at first.
Those positive results often faded. In some cases, the outcomes flattened out or disappeared entirely.
Why?
Based on a small set of interviews and anecdotal observations, I came to a conclusion that has stuck with me ever since: in many cases, the early success wasn’t primarily a blended learning effect. It was a selection effect—of teachers.
The most motivated teachers were selecting themselves into the blended learning initiatives. A school would ask for teachers interested in volunteering to participate. The teachers who raised their hands were more interested in blended learning than the average teacher. Arguably, many of them were also stronger than average in other ways: more reflective, more open to experimentation, and more willing to put in extra effort to make something new work.
Those teachers approached the pilot with a high level of energy and intentionality. Students benefited. Outcomes improved.
Then the program expanded.
When blended learning moved from a voluntary pilot to a required implementation, across an entire grade level, subject area, or school, the selection effect disappeared. The teacher population now looked much more like the overall distribution.
In many cases, blended learning itself wasn’t the primary driver of early success. The selection of highly motivated teachers played a much larger role.
Outcomes appear worse in many online schools due to selection effects
A different version of this dynamic becomes even more visible in online schools.
Online educators know this well. Their students are often highly mobile, credit-deficient, parenting or working full-time, medically fragile, or re-engaging after prior disengagement. Many are navigating housing instability or family disruption.
Online schools don’t create these conditions; they inherit them. Yet when graduation rates, proficiency scores, or accountability labels are released, those selection effects often disappear from the narrative. Outcomes are reported as if student populations were interchangeable. They aren’t.
Choice tends to amplify selection effects
One reason selection effects are going to matter even more in the coming years is the steady expansion of choice in K–12 education.
When families have more options, they sort. Sometimes intentionally, sometimes implicitly. That sorting may reflect academic preferences, schedules, instructional philosophy, student needs, or a family’s capacity to navigate options. Regardless of motivation, the result is the same: student populations become less interchangeable.
We’re already seeing early versions of this dynamic in schools that market themselves as “AI-driven.” In at least some cases, outcomes cited as evidence of AI’s effectiveness are influenced as much by who is opting into the school as by the instructional tools themselves. Families choosing these environments tend to be more engaged and more willing to experiment, which changes the starting conditions before any technology is applied.
The same pattern will almost certainly play out at a much larger scale as education savings accounts and private school voucher programs continue to expand.
Choice increases sorting, and sorting amplifies selection effects.
This isn’t an argument against choice. But it does make outcome comparisons harder—and more fragile—than most accountability systems are designed to handle.
Why current accountability systems are no longer fit for purpose
Selection effects help explain why accountability systems in K–12 education have struggled to adapt.
At some level, selection effects are acknowledged. Policymakers and researchers will often note differences in student populations or say that context matters. But those acknowledgements rarely translate into accountability frameworks that fully adjust for selection effects. Part of the challenge is technical. Fair comparisons across schools serving very different students require better data, more nuanced metrics, and a greater tolerance for complexity than most systems currently allow.
But part of the challenge is structural. For decades, accountability has been built on the assumption that schools are broadly comparable units—that with enough controls, outcomes can be used to rank, label, and intervene. That assumption becomes increasingly untenable in a system defined by choice, mobility, and specialization.
As selection effects grow stronger, accountability systems face questions they are poorly equipped to answer:
- Are we measuring school quality, or student composition?
- Are we discouraging schools from serving students who most need alternative options?
In a more choice-rich environment, accountability systems designed for a largely uniform student population are no longer fit for purpose.
What current accountability systems produce
As accountability systems fail to grapple seriously with selection effects, the result isn’t just unfair labels, it’s also distorted incentives.
Schools and programs that serve students with more complex needs will continue to look weaker on paper, even when they are doing effective and necessary work. Meanwhile, schools better positioned to attract already-advantaged students will continue to look strong, regardless of instructional quality.
Selection effects aren’t a footnote to modern education systems. They are a defining feature. Treating them as a secondary concern doesn’t make accountability simpler; it makes it less honest.
We’re going to be thinking more about this topic, and what DLAC can do about it, across 2026, and welcome your thoughts.
We welcome your comments in the DLAC Community Portal’s Blog Discussion Group. Scroll down to the bottom of the page to join the conversation.