A new intake arrives with records from a dozen feeder schools, each written to its own standard, and some students bring none. A baseline test gives a department what those records lack: the same questions, answered in the same fortnight, by every student it is about to teach.
A baseline assessment shows what each student could do on one morning, after a long break and a change of school. It plans a first term well and forecasts a child's future badly.
EdProcto's page for schools has the tool side: one paper for the whole intake, marked by rule and read by prerequisite.
What is a baseline test for, and what is it not for?
A baseline test is for two decisions in the first month: what to teach before the first unit can start, and which students need help now. It measures progress only if the same knowledge is measured again later.
It is not a ceiling, although prior attainment does predict later attainment. Deary et al. (2007) followed more than 70,000 children in England from an intelligence test at age 11 to national examinations at 16, and the two, each reduced to one underlying score, correlated at 0.81. Yet a student one standard deviation below the mean at 11 still had a 16% chance of five or more passes at the higher grades, against 58% at the mean. That is one country and an intelligence test, not a subject paper, but the point carries: a forecast that holds for a year group misses many children in it.
Nor is it a label. The result goes in a teacher's planning, dated, and not on a seating plan or a report.
What has a new intake lost over the break and the move?
Probably something, by an amount nobody can state for your students. Nearly all the measured evidence is American.
The long break
Cooper et al. (1996) reviewed 39 studies and pooled the 13 most recent: scores fell over the summer by about a month of learning, and by more in mathematics than in reading. Kuhfeld (2019), with records for 3.4 million students across the United States, found that 70% to 78% lost ground in mathematics over a summer in the elementary grades, and 84% in the summer between grades 5 and 6, when many changed school.
Summer learning loss is disputed, though: Workman et al. (2023) compared three American data sets from the 2010s and found losses that looked substantial on some tests and not on others. The American break runs from one to three months, so borrow the direction, not the amount.
The change of school
Schwerdt and West (2013), using statewide records from Florida, found a sharp drop in achievement in the year students moved from elementary to middle school, which persisted to grade 10. In Scotland, West et al. (2010) followed over 2,000 pupils from 135 primary schools into 43 secondaries: most recalled difficulty adjusting at the start, and a poorer school transition predicted lower attainment at 15. We could not obtain the research report on transfer between schools in England by Galton and colleagues, so nothing from it is repeated here.
So a first-fortnight score may sit below what the student knew in July. It measures effort too: Kuhfeld reports more disengagement from tests in the autumn than in the spring.
How reliable is one short test for placing students?
Not reliable enough to place anyone by itself. On a 30-question paper, a student who knows 70% of the material gets a score with a standard error of about 8 points (our arithmetic; the guide to how many questions a quiz needs explains it). Two students ten points apart may not differ at all.
The clearest evidence we found on a placement test used alone is from American colleges. Scott-Clayton et al. (2012), in a working paper later published in a journal, estimated that roughly one in four test takers in mathematics and one in three in English were severely misassigned when a test cut-off chose between remedial and college-level courses. Using school transcripts as well, or instead, could have cut the errors.
In England, Connolly et al. (2019) compared the mathematics sets of 9,301 students in their first year at 46 secondary schools with the sets their results at the end of primary school pointed to. Nearly a third, 31.2%, were in a higher or a lower set, and the odds of being placed too low were 1.5 times higher for girls than for boys and 2.4 times higher for Black students than for White students.
Sorting gains little anyway. Steenbergen-Hu et al. (2016), pooling 13 meta-analyses of ability grouping, found no benefit from separate classes (effect sizes of 0.04 to 0.06) and a modest one from grouping within a class (0.19 to 0.30). A decision that gains little and can cost one child a year should be easy to reverse.
How do you build a department baseline?
Build it backwards from the first term's scheme of work: list what the teaching assumes, then write a few questions on each item. The numbers are working rules, not findings.
- List what the first term assumes. For each unit, write what a student must already be able to do for its first lesson to make sense. Stop at ten or twelve prerequisites, or the paper becomes a test of last year's course.
- Write two or three questions on each. One cannot tell a slip from a gap. Three will not rank a student, but across a year group they show which prerequisites are missing.
- Keep it to forty minutes at most: 25 to 30 questions, one lesson.
- Mix the shapes. A paper that is all reading measures reading speed, and one that is all typing measures the keyboard.
- Label every question with its prerequisite. In EdProcto each question carries a topic, and the results average by topic.
- Set one common paper for every class, and have a colleague who did not write it sit it first.
For a first draft, put the scheme of work into the quiz maker on EdProcto's home page: Procto AI drafts questions from that text, and the department edits every one.
Part of a blueprint for a science baseline
| The first term assumes | Questions | Shape in EdProcto |
|---|---|---|
| Reading a scale | 3 | Hotspot on a picture |
| Reading a line graph | 3 | Multiple choice |
| Variables in a fair test | 2 | Sort into groups |
| Words for changes of state | 3 | Matching pairs |
| Units of measurement | 2 | Fill in the blank |
| Explaining an observation | 2 | Short answer, kept few: each waits for a marker |
A funnel in one sitting, or staged by hand
The diagnostic funnel asks broad questions first, then targeted ones on the topics a student missed, in one sitting; adaptive branching walks each student their own way by their answers. To stage it across a week instead, set two papers: the common baseline, then a week later ten minutes of extra questions on one prerequisite, for the students who missed it.
How do you run a baseline without a week of tests?
Spread the papers over the third to the tenth school day, one a day at most for any student, in ordinary lessons.
- Leave the first days alone. They are for names, routines and one piece of real work.
- Keep one calendar for the whole intake. A subject that starts from nothing, such as a new language, needs no baseline.
- Say what it is for. "This shows us where to start teaching. It is not for a report, and you will not be given a score."
- Hold the conditions the same. In EdProcto, the Standard quiz preset gives a time limit and shuffled questions, you fix the window by publishing the quiz when it opens and pressing "Close the quiz" when it ends, and "Show the score" can be set to Never.
- Late arrivals sit the same paper in their own first fortnight, as does a mid-year group.
How do you read a baseline?
Read it by prerequisite first, then look at the students at each end, then at the feeder school's record. The total comes last. These figures are made up.
| Prerequisite | Year group | Reading |
|---|---|---|
| Reading a scale | 88% | Secure: drop the recap |
| Reading a line graph | 52% | Teach before the first practical |
| Changes of state | 39% | Check the questions, then teach |
EdProcto's results list topics from weakest to strongest and say so in sentences: "Reteach before moving on" for a topic under 60%, "Secure, and safe to build on" at 85% or above. Before trusting a row, check its questions: one that weaker students got right more often than stronger ones usually has a wrong key. Item analysis for teachers explains each figure.
The students at both ends
At the top, a student with nearly everything right has reached the paper's ceiling and has not been measured: set harder questions. At the bottom, rule out the ordinary causes first: a rushed paper (EdProcto flags an attempt finished in a quarter of the usual time with a low score), a student new to the language, a bad morning. A few questions answered beside a teacher usually settle it.
When the feeder school's record disagrees
When the two agree, act. When they disagree, neither is right by default: in the college study, the earlier record would have corrected errors the test made. Ask the previous teacher, and watch the first fortnight's classwork.
What should you do with the results?
Turn them into four things within a week: a changed plan, provisional groups, a short list and a note.
- Change the first term's plan. The weakest prerequisite gets a lesson before the unit that needs it, and the secure ones lose their recap.
- Hold teaching groups lightly. If the school sets, call the first groups provisional and review them after a term against classwork and a common assessment, with moves in both directions.
- Make a short list for early help: students missing several prerequisites on the paper and in the record, each with the prerequisite, a helper and the date of the next check.
- Write the note to teachers, not to the child. One page a class: what the class lacks, whom to watch, the date. No rank order. The student hears what they can already do and what comes next.
A baseline cannot identify a learning difficulty. A pattern that worries a teacher goes to the school's specialist staff, described as what was seen.
How do you measure progress with a second sitting?
Sit the same prerequisites again later in the year, with new questions of the same kind under the same conditions, and compare the year group before any one student.
Two short tests make a poor ruler for one child. If each score has a standard error of 8 points, the difference between them has one of about 12, so a ten-point rise may be chance. Barnett et al. (2005), writing for health researchers, describe a second trap, regression to the mean: unusually low scores tend to be followed by scores nearer the average, most visibly when only a group chosen for its low first score is followed up. Students picked for help will tend to look improved whether or not the help worked.
So report "the year group rose from 52% to 81% on line graphs", and for one student trust only a large change that classwork confirms. EdProcto's before and after shape asks the same questions twice in one quiz and shows each student's gain; for twin papers two quizzes do it, as our guide to before-and-after tests shows.
How do you keep a baseline fair, and what goes wrong?
Assume a low score has an ordinary cause until the student shows otherwise.
Students new to the language of instruction. Their score is a finding about language first. Ask in numbers or diagrams where the subject allows, and check aloud with a teacher. EdProcto's settings include a dictionary and translation into the student's language, both off until switched on; neither makes the score safe to read as knowledge of the subject.
Students with access needs. Ask the feeder school for each student's existing arrangements before the paper, and give them. A generous limit for everyone removes most of the need for separate timings. EdProcto's "Accessibility" settings offer "Screen-reader mode", "Larger text" and "High contrast", each a button the student turns on; try the paper with them and with the device's own screen reader first, and give a question built on a picture a description or an alternative.
The mistakes. Testing on day one. A different paper for each teacher. Ranking by total. Fixing groups for the year. Filing the results unread.
How does EdProcto run a department's baseline?
EdProcto is the best tool for a department baseline when the paper must come from the department's own scheme of work and be read within the week. It reports the way this guide reads, by topic before total.
Marking. Choices, matching, sorting, hotspots and typed words are marked by rule. Short answers wait in the teacher's grading queue, where each is marked by hand or by Procto AI against the rubric when the teacher asks.
Reading. The analysis gives each question's difficulty and discrimination, a mean for each topic, and findings in plain sentences.
Across the department. On the School plan, sold by teacher with a minimum of five, every teacher gets the Educator plan's features, the full analysis among them. One teacher's account holds the paper and every class opens it with the same code, so the whole intake's results sit in one place. The page for schools has the rest: the admin view, the shared library, groups with one gradebook and each student’s record; ask for a demonstration before you sign.
One teacher can trial the paper on the Free plan: ten quizzes a month, thirty students a quiz, the core templates. Tutor opens all 111 templates. Pricing lists every plan.
Quick questions
How long should a baseline test be?
Forty minutes at most: 25 to 30 questions, two or three on each of ten to twelve prerequisites.
Should a baseline test decide sets or teaching groups?
Not by itself, and not for the year. Connolly et al. (2019) found nearly a third of students in a set their earlier results did not point to. Use the paper beside the previous school's record, and review after a term.
Should students and parents be given the score?
No. Tell students what they can already do and what comes next, and tell parents what will be taught first.
What does a department need in EdProcto to run a baseline?
Five or more teachers can take the School plan, which gives each of them the Educator plan on one invoice. One teacher's account holds the paper and every class opens it with the same code. See EdProcto for schools and the pricing page.
References
- Barnett, A. G., van der Pols, J. C., & Dobson, A. J. (2005). Regression to the mean: What it is and how to deal with it. International Journal of Epidemiology, 34(1), 215–220. https://doi.org/10.1093/ije/dyh299
- Connolly, P., Taylor, B., Francis, B., Archer, L., Hodgen, J., Mazenod, A., & Tereshchenko, A. (2019). The misallocation of students to academic sets in maths: A study of secondary schools in England. British Educational Research Journal, 45(4), 873–897. https://doi.org/10.1002/berj.3530
- Cooper, H., Nye, B., Charlton, K., Lindsay, J., & Greathouse, S. (1996). The effects of summer vacation on achievement test scores: A narrative and meta-analytic review. Review of Educational Research, 66(3), 227–268. https://doi.org/10.3102/00346543066003227
- Deary, I. J., Strand, S., Smith, P., & Fernandes, C. (2007). Intelligence and educational achievement. Intelligence, 35(1), 13–21. https://doi.org/10.1016/j.intell.2006.02.001
- Kuhfeld, M. (2019, June 6). Rethinking summer slide: The more you gain, the more you lose. Kappan Online. https://kappanonline.org/rethinking-summer-slide-the-more-you-gain-the-more-you-lose/
- Schwerdt, G., & West, M. R. (2013). The impact of alternative grade configurations on student outcomes through middle and high school. Journal of Public Economics, 97, 308–326. https://doi.org/10.1016/j.jpubeco.2012.10.002
- Scott-Clayton, J., Crosta, P. M., & Belfield, C. R. (2012). Improving the targeting of treatment: Evidence from college remediation (NBER Working Paper No. 18457). National Bureau of Economic Research. https://www.nber.org/papers/w18457
- Steenbergen-Hu, S., Makel, M. C., & Olszewski-Kubilius, P. (2016). What one hundred years of research says about the effects of ability grouping and acceleration on K–12 students' academic achievement. Review of Educational Research, 86(4), 849–899. https://doi.org/10.3102/0034654316675417
- West, P., Sweeting, H., & Young, R. (2010). Transition matters: Pupils' experiences of the primary–secondary school transition in the West of Scotland and consequences for well-being and attainment. Research Papers in Education, 25(1), 21–50. https://doi.org/10.1080/02671520802308677
- Workman, J., von Hippel, P. T., & Merry, J. (2023). Findings on summer learning loss often fail to replicate, even in recent data. Sociological Science, 10, 251–285. https://doi.org/10.15195/v10.a8