By Aaron Hepburn
One grade level in a district we work with finished last spring at 88 percent proficient in English language arts. That is fifteen points above the state average and the strongest result in the building. In the same year, in the same grade, the median student growth percentile (SGP) was 39.5, where 50 is typical.
Both numbers are accurate. Put together, they describe a group of students who arrived already ahead and then moved more slowly through the year than academically similar students elsewhere in the state. Read the first number by itself and this is the district's best story. Read both and it becomes a question worth answering before that cohort reaches high school.
In the districts we work with, a grade level like this one is common. What is far less common is a data review built to surface it.
This is the first of two posts. This one is about how to read the data. The second, “Reading Data Without anyone Getting Defensive,” is about what happens in the room when you do.
Proficiency and SGP raise and answer different questions about the learning happening in a district.
Proficiency is a status measure. It answers one question: what share of students met the standard at the moment the test was given? It is a snapshot. It says nothing about the distance any student traveled to get there.
SGP answers the other question. For each student, the SGP model finds other students across the state who scored about the same on the same test in prior years, then reports where that student's current score falls within that group. A student at the 60th growth percentile grew more than 60 percent of the students who started in the same academic place. Roll the individual results up and take the middle value, and you get a median SGP for a grade, a school, or a district. Damian Betebenner introduced the model in 2009; Castellano and Ho's practitioner's guide for the Council of Chief State School Officers is the plain-language companion. It began in Colorado and is now used in more than two dozen states, which is why SGP appears on state report cards that otherwise share very little.
Three things to hold onto when reading SGP:
In everyday terms: proficiency asks whether a ninth grader can do ninth grade work. SGP asks how far that student moved this year compared with students across the state who started the year in the same academic place. The two answers come apart more often than people expect. A student can fall short of proficient and post an SGP of 70, because the comparison is against students who started where she started, not against the standard. A student can clear the proficiency bar and post an SGP of 30.
Sean Reardon at Stanford built the Stanford Education Data Archive from roughly 300 million test scores covering every public school district in the country. His finding: average third-grade scores in a district track closely with the learning opportunities available before and during early elementary school, which are themselves closely tied to family income and parent education. Average growth from third to eighth grade barely relates to those third-grade scores at all. Growth in many higher-poverty districts outpaced growth in wealthier ones.
Reardon's summary is worth quoting: “Poverty clearly does not determine the quality of a school system.”
The practical consequence is direct. A district's proficiency rate is substantially a statement about who its students are. Its growth rate is much closer to a statement about what happens to them once they arrive. If your improvement plan is built entirely on proficiency, it is partly built on demographics you do not control.
Kane and Staiger, writing in the Journal of Economic Perspectives in 2002, pointed out something that has never stopped being true: the median elementary school in the United States has about 69 children per grade level. At that size, a single year's grade-level result carries a great deal of statistical noise. A twenty-point swing in one grade can be a real instructional change, or it can be thirty different families moving in and out.
Growth percentiles are not immune. There is a serious published critique, argued most directly by Stephen Sireci, Craig Wells, and Lisa Keller at the University of Massachusetts, that individual SGPs carry enough measurement error that a student retaking the same test could land in a noticeably different place. Researchers who study how principals actually use growth data have found similar interpretation problems in practice.
We take that critique seriously rather than arguing with it, because it points at the same discipline: read the trend, not the dot. Three years of a measure tells you something a single year cannot. One year is an anecdote with a decimal point.
Two notes before the charts:
Here is the same district's most recent year, with proficiency on the horizontal axis and median growth on the vertical. The horizontal line is typical growth. The vertical line is the state proficiency average.
The grade levels separate immediately, and they separate into four regions that ask four different questions. The idea that these four patterns point to distinct underlying causes comes from the Cohesive Learning framework developed by We Do Instruction, also known as Cohesive Schools, which names three interdependent factors behind whether learning accumulates: access to core instruction, the strength and rigor of core content, and whether learning gets solidified over time. The framework draws on Daniel Willingham (2006) on background knowledge, Bjork and Bjork (2011) on the conditions that produce durable learning, and Richard Elmore's instructional core (2009). We use it as one lens among several when we sit down with a district's data.
Higher proficiency, stronger growth. Grades 4, 7, 9, 10 and 11. Students are ahead and continuing to move. The question here is protection. What is producing this result, and what would put it at risk? That is harder to answer than it sounds, because nobody schedules a meeting about the thing that is working.
Higher proficiency, slower growth. Grade 8, at 88 percent proficient and a median SGP of 39.5. This is the quadrant that a proficiency-only review will never show you. The question here is rigor. Is the work demanding enough for students who already arrive capable? Look at text complexity, sustained writing, and whether students are asked to transfer and reason rather than keep pace with coverage.
Lower proficiency, slower growth. Grades 5 and 6. The question here is access. Can every student engage with grade-level content, or are gaps in background knowledge and academic vocabulary putting the material out of reach for part of the room?
Lower proficiency, stronger growth. Empty in this district, which is itself information. When grades land here, students are moving faster than their academic peers and the gains have not yet accumulated into proficiency. The question here is whether the gains hold. Is learning being solidified and revisited over time, or does each unit start over? Part 2 of this series shows a second district with three grade levels sitting in exactly this position.
Everything to this point comes from a district testing above its state average, which makes it easy to read this as an exercise for people who already have good news to report. The reverse case is where the chart earns its keep.
Go back to the two ideas from earlier in this piece. Proficiency largely reflects who your students are. Growth largely reflects what happens to them once they arrive. Follow that through, and a hypothetical district at 55 percent proficient with grade levels posting a median SGP of 60 has hard evidence of instructional work that no proficiency report will ever credit it for. That evidence is worth having, for the staff doing the work and for the board asking why the number is not higher yet.
The chart cuts both ways. High scores can hide a real problem. Low scores can hide real momentum. Neither one surfaces until you plot both measures on the same page.
The single-year chart tells you where each grade sits. It does not tell you where each grade is heading. Add three years and each grade level grows a tail: a line running from where it sat three cycles back to where it sits now.
The hollow marker is where the grade level sat three cycles back. The solid marker is the most recent year. The line between them is the story.
Grade 8 is now the most interesting point on the page. Its proficiency has climbed steadily, 78 to 81 to 88 percent. Its median SGP has moved the other way: the 54th percentile, then the 61st, then 39.5. Proficiency rising while growth rises and then falls sharply is a specific signature, and it is not a fluke year. It suggests a grade level absorbing capable students and not adding much to them.
Grade 6 tells a different story, quieter, and pointing at access rather than rigor. Proficiency 77, 72, 72. Median SGP at the 46th percentile, then the 45th, then the 33rd. Nothing dramatic in any single year. A clear drift across three.
Grades 10 and 11 barely move, which is exactly what you want to see in the strong quadrant. Stability there is a finding in its own right, not an absence of one.
You can do this in a spreadsheet in under an hour. No new platform required.
Two cautions worth stating plainly. Do not put this chart in front of a subgroup with fewer than twenty or thirty students and treat the result as a trend; small cohorts swing hard, and the honest read is that you cannot tell yet. And do not let this chart anywhere near teacher evaluation. It is a diagnostic for a system, at a grain size of a grade level and a year. Pointed at individuals it produces defensiveness and bad decisions, and it stops being useful for the thing it is good at.
The chart is not the work. It is the thing that makes the work discussable.
In Part 2, “Reading Data Without Getting Defensive,” I will walk through what happens after the chart goes up on the screen: how leadership teams have used this view to get past the argument between “our scores are up” and “our system feels fragmented,” why reading three years instead of one takes the personal sting out of the conversation, and how four regions on a chart turned into a sequenced set of priorities in two different districts.
If you want to try this with your own numbers and would rather not do it alone, we run this analysis with district leadership teams as a starting point for planning. Reach out and we will look at your data together.
Betebenner, D. W. (2009). Norm- and criterion-referenced student growth. Educational Measurement: Issues and Practice, 28(4), 42–51.
Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher, R. W. Pew, L. M. Hough, & J. R. Pomerantz (Eds.), Psychology and the real world: Essays illustrating fundamental contributions to society (pp. 55–64). Worth Publishers.
Castellano, K. E., & Ho, A. D. (2013). A Practitioner's Guide to Growth Models. Council of Chief State School Officers.
City, E. A., Elmore, R. F., Fiarman, S. E., & Teitel, L. (2009). Instructional Rounds in Education: A Network Approach to Improving Teaching and Learning. Harvard Education Press.
Clauser, A. L., Keller, L. A., & McDermott, K. A. (2016). Principals' uses and interpretations of student growth percentile data. Journal of School Leadership, 26(1), 6–33.
Cohesive Learning framework, including Academic Momentum and the Three Factors: We Do Instruction (Cohesive Schools), wedoinstruction.com. Used with attribution.
Kane, T. J., & Staiger, D. O. (2002). The promise and pitfalls of using imprecise school accountability measures. Journal of Economic Perspectives, 16(4), 91–114.
Reardon, S. F., and colleagues. Stanford Education Data Archive and the Educational Opportunity Project, Stanford University.
Sireci, S. G., Wells, C. S., & Keller, L. A. Why we should abandon student growth percentiles. Center for Educational Assessment, University of Massachusetts Amherst.
Spector, C. (2017, December 5). Poverty, early test scores do not determine the quality of a school system, research shows. Stanford Institute for Economic Policy Research. (Source of the Reardon quotation.)
Willingham, D. T. (2006). How knowledge helps. American Educator, 30(1), 30–37.
Aaron Hepburn is Co-Founder of Compass PD, LLC, where he leads the analytical data reviews Compass PD conducts with partner districts.