School teacher to data analyst: item analysis from a marks register

Several beliefs come up again and again when teachers consider this move. Here is what each one looks like against the evidence you can check yourself.

"I've never worked with data"

What the evidence says: you have, just not in a tool called a database. A marks register is a table: one row per student, one column per assessment, with a total and a grade. An attendance register is a daily event log. If you have ever entered question-wise marks for a unit test into a spreadsheet so the school could see which chapters the class struggled with, you have done the first half of an analysis.

What you probably have not done is store that data in a way that survives more than one term, join it to another table (attendance to marks, say), or check it for errors systematically. Those are learnable, and they are exactly what an analyst's first months involve.

"Teaching doesn't count as business experience"

What the evidence says: it depends on how you describe it. Parent-teacher meetings are stakeholder meetings. A remedial plan for a group of students is an intervention with a measurable outcome. Deciding which of four sections needs extra periods before board exams is resource allocation based on evidence.

The gap is vocabulary and metrics. Businesses talk about conversion, retention, churn and revenue. Retention is the same idea as tracking which students stay enrolled in an optional coaching programme; churn is the students who drop out. Learning to translate is part of preparing, and it is why your first project should use education data you understand before you move to a sales or operations dataset.

"I'll have to start from zero"

What the evidence says: you start from zero on tools, not on reasoning. Here is a project that uses reasoning you already have.

Item analysis on a synthetic unit test

Take a ten-student, five-question test (all data invented). For each question, compute two numbers:

• Difficulty index — the share of students who got it right. Near 1.0 is easy, near 0 is hard.
• Discrimination index — the share correct among the top-scoring group minus the share correct among the bottom-scoring group. A question that strong students get right and weak students get wrong discriminates well; a question with a negative value is suspicious.

import pandas as pd

# Synthetic question-wise results: 1 = correct, 0 = wrong
data = {
    "student": ["S01","S02","S03","S04","S05","S06","S07","S08","S09","S10"],
    "Q1": [1,1,1,1,1,1,1,1,0,1],
    "Q2": [1,1,1,1,0,1,0,0,0,0],
    "Q3": [1,1,1,0,1,0,1,0,0,0],
    "Q4": [0,0,0,1,0,1,0,1,1,1],
    "Q5": [1,1,0,1,0,0,0,0,0,0],
}
df = pd.DataFrame(data).set_index("student")
df["total"] = df.sum(axis=1)

ranked = df.sort_values("total", ascending=False)
top, bottom = ranked.head(3), ranked.tail(3)

questions = ["Q1","Q2","Q3","Q4","Q5"]
report = pd.DataFrame({
    "difficulty": df[questions].mean(),
    "discrimination": top[questions].mean() - bottom[questions].mean(),
}).round(2)
print(report)

Output:

    difficulty  discrimination
Q1         0.9            0.33
Q2         0.5            1.00
Q3         0.5            0.67
Q4         0.5           -0.67
Q5         0.3            1.00

Read it the way you would read a class's results. Q1 was easy for nearly everyone, so it tells you little. Q2 separated strong and weak students cleanly. Q4 has a negative discrimination — the weaker students got it right and the stronger ones did not. In real tests that usually means an ambiguous question, a wrong answer key or a trick that rewards guessing. Your recommendation: check the answer key for Q4 first, then rewrite it.

That recommendation — spot the anomaly, name the likely cause, say what to check before acting — is analyst work. For a fuller project, add a second test and attendance, and ask whether students with lower attendance lost marks on specific chapters.

"A course will teach me everything I need"

What the evidence says: a course gives you tools and structured practice; the portfolio and the explanation are still yours to produce. The Data Analyst programme takes you through Excel, SQL, Power BI and Python with graded deliverables, but the item-analysis project works best when you choose which anomaly to investigate.

The specific gaps for most teachers:

• SQL — grouping, joining two registers, filtering by date.
• Spreadsheet modelling beyond marks — lookups across sheets, pivot tables on thousands of rows.
• Business metrics — learn one domain's language well, such as retail or edtech operations.
• Speed with messy data — real exports have duplicates and misspelled names; school registers are usually cleaner than company data.

"Only an edtech company will look at me"

What the evidence says: edtech is a natural first target because your domain knowledge is directly relevant, but it is not the only one. Aim first for titles such as junior data analyst, MIS analyst or academic operations analyst, and read each posting for whether it asks for SQL and dashboards rather than only report formatting.

The answer to "why are you leaving teaching?"

I'm not leaving the part I enjoyed most, which was working out why a group of students was struggling and what to change. I realised I was doing that with spreadsheets every term, so I learned SQL and Python to do it properly. In my portfolio I ran an item analysis on a test and found one question where weaker students scored better than stronger ones — which pointed to an ambiguous question, not a learning gap. I'd like to do that kind of diagnosis on business data, where the same question comes up: is the problem the people, or the measurement?

Before applying, the free skill check shows which of those tools your current CV actually evidences.