AdvancedLesson 30 of 307 min read

Capstone: Should the Workshop Open a Second Evening?

Answer the first question of the course with everything you have learned: a measure, a range, seasons, a decision and its doubts.

In this lesson you will

  • Plan an analysis from question to decision
  • Use a range and a seasonal breakdown together
  • Write a recommendation with its doubts
  • Review the habits of the whole course

About 34 minutes

Back to the first question

In lesson 1 of this course, Leila asked a question: “Should we open a second evening?” It was a good question, and the first answer, from six weeks of tiny numbers, was a hesitant no. Now you have the tools to answer it properly. The capstone does not teach anything new. It uses what you already know on one question from start to finish, and it shows what a finished piece of analysis looks like: a clear question, a trusted table, a fair measure, a range, a decision, and a list of doubts.

The plan

  1. Question. Should Spoke and Co open an extra evening, given a made-up capacity of six visits a session?
  2. Measure. The share of sessions at or over capacity, and the visits per session. The measure is written in a sentence before any calculation, as in lesson 11.
  3. Data. The ten-month record of session counts, checked with the habits of lessons 21 and 22.
  4. Analysis. Overall rate, a range for it, the rate by month, and the number of visits that exceeded capacity.
  5. Decision. A recommendation with its doubts.
import pandas as pd, numpy as np
ts = pd.read_csv("sessions.csv", parse_dates=["date"])
cap = 6
print("question: second evening? (made-up capacity", cap, "visits a session)")
n = len(ts)
print("sessions:", n, "| mean visits:", round(ts["visits"].mean(), 2))
full = (ts["visits"] >= cap).sum()
p = full / n
z = 1.96
centre = (p + z * z / (2 * n)) / (1 + z * z / n)
half = z * ((p * (1 - p) / n + z * z / (4 * n * n)) ** 0.5) / (1 + z * z / n)
print("sessions at or over capacity:", full, "of", n, "=", round(p * 100, 1), "percent")
print("range:", round((centre - half) * 100), "to", round((centre + half) * 100), "percent")
ts["month"] = ts["date"].dt.month
by = ts.groupby("month")["visits"].agg(["mean", lambda s: (s >= cap).mean()])
by.columns = ["mean", "share"]
busy = by[by["share"] >= 0.5]
for m, row in by.iterrows():
    print(pd.Timestamp(2026, m, 1).strftime("%b") + ": " + str(round(row["share"] * 100)) + " percent full")
print("months with half the sessions full:", [pd.Timestamp(2026, m, 1).strftime("%b") for m in busy.index])
rng = np.random.default_rng(30)
v = ts["visits"].to_numpy()
boot = [rng.choice(v, size=n, replace=True).mean() for _ in range(5000)]
lo, hi = np.percentile(boot, [2.5, 97.5])
print("mean visits, 95 percent range:", round(lo, 2), "to", round(hi, 2))
extra = (np.clip(v - cap, 0, None)).sum()
print("visits above capacity in all sessions:", int(extra))
print("recommendation: trial evening in busy months only" if p < 0.5 and len(busy) else "recommendation: see figures")
checks = [("sessions counted", n == len(v)), ("visits not negative", (v >= 0).all()), ("months 1 to 10", set(ts["month"]) == set(range(1, 11)))]
for name, ok in checks:
    print(("PASS " if ok else "FAIL ") + name)
question: second evening? (made-up capacity 6 visits a session)
sessions: 85 | mean visits: 4.16
sessions at or over capacity: 24 of 85 = 28.2 percent
range: 20 to 39 percent
Jan: 0 percent full
Feb: 0 percent full
Mar: 0 percent full
Apr: 12 percent full
May: 67 percent full
Jun: 44 percent full
Jul: 25 percent full
Aug: 0 percent full
Sep: 67 percent full
Oct: 71 percent full
months with half the sessions full: ['May', 'Sep', 'Oct']
mean visits, 95 percent range: 3.81 to 4.53
visits above capacity in all sessions: 8
recommendation: trial evening in busy months only
PASS sessions counted
PASS visits not negative
PASS months 1 to 10

Over 85 sessions, the mean is 4.16 visits, comfortably under capacity. The bootstrap range for that mean runs from 3.81 to 4.53, so even the high end is well below six. The share of sessions at or over capacity is 24 of 85, or 28.2 percent, with a range of 20 to 39 percent. That is a use of the Wilson range from lesson 15: with eighty-five sessions the share is known only to within about ten points either way.

The monthly breakdown tells the real story. It is the seasonal pattern from lesson 14. In January to March and in August, no session reached capacity at all. In May, September and October, more than half did. The months in which at least half the sessions were full are [‘May’, ‘Sep’, ‘Oct’]. Notice that October is partial, with seven sessions so far, so the October figure is a rate and not a total, and it comes from few sessions. The most telling number may be the last: across all 85 sessions, only 8 visits were above capacity. People who could not be served were rare, even in the busy months.

The recommendation, with its doubts

A decision needs a sentence, not a table. Here is one: do not add a permanent second evening. Run a trial extra session in the busy months, and review it after a season. The reasons follow directly from the figures. Demand is below capacity most of the year. A few months fill up often, and extra capacity would be used in those months and wasted in the others. And the cost of being wrong is asymmetric: a trial can be undone, and a permanent evening is hard to cancel if volunteers have rearranged their lives.

Then the doubts, which are part of the answer and not an apology:

  • The capacity figure is invented. If the real limit is five visits, the picture changes. Leila should check it with the mechanics.
  • The record is less than a year. We have not seen last winter, so we do not know whether the autumn rise repeats.
  • Counts show visits, not turned-away customers. People who saw a queue and left would not be in the data. Ask the volunteers whether that happens.
  • Volunteers are not hours on a spreadsheet. A second evening depends on people, and that is a question for conversation, not for data.

The checks, and what you have learned

The last lines of the output show three checks passing: the number of sessions agrees with the number of values, no count is negative, and every month from one to ten is present. They are small, and they are the kind that stop embarrassing mistakes. Behind them sit the larger habits of this course: define the measure, trust the table only after checks, give a range with every estimate, count your comparisons, remember who is missing, and write the answer in a sentence that someone can act on.

You have learned to ask a question, clean a table, summarise and group, find trends and seasons, join tables, estimate uncertainty, test a gap, fit a line, code comments, write reports, test your code, handle odd values and gaps, forecast, share data carefully, price the work and plan a project. None of that makes you a statistician, and you do not have to be one. It makes you someone who can look at numbers and say what they show, how sure you can be, and what they cannot say. That is what data analysis is for.

Common mistake: no conclusion

Presenting a table of results and letting the reader decide what it means. A finished analysis ends in a decision or a recommendation, with the doubts beside it.

Made-up workshop

The session counts, the capacity and the workshop are made up. The recommendation describes this invented workshop only. The method is the point: the same steps would serve a real one.

Choose a real question from your own life or work. Write the five steps of the plan in one line each, then list the three biggest doubts before you collect any data.

Practice for this lesson

Check what you learned

Five quick questions. Pick an answer and you will see straight away whether it is right and why.

Why does the capstone repeat the first question of the course?

  1. To answer it again with better tools and honest doubts
  2. Because the first answer was wrong in every way
  3. To avoid new data
  4. To skip the cleaning

The tools of the course turn a hesitant no into a reasoned recommendation.

What did the monthly breakdown show?

  1. Every month is equally busy
  2. Sessions are often full in a few busy months and rarely in the rest
  3. There were no sessions in May
  4. Capacity is exactly six everywhere

May, September and October were the busy months.

Why recommend a trial session instead of a permanent evening?

  1. Trials are cheaper than data
  2. Permanent evenings are illegal
  3. The data is perfect
  4. A trial can be undone, and demand is below capacity most of the year

The cost of being wrong is not the same both ways.

Which doubt should a good report mention?

  1. That the report took a long time
  2. The capacity figure is invented and the record is under a year
  3. That the data was large
  4. That the charts were plain

Doubts are part of the answer, not an apology.

What is the purpose of the final checks in the output?

  1. To change the results
  2. To make the report longer
  3. To replace the decision
  4. To stop small embarrassing mistakes

They are small, and they protect the whole analysis.

Summary

  • Define the measure before calculating.
  • Combine an overall figure, a range and a seasonal view.
  • End with a decision and a list of doubts.
  • Check the work, and remember who and what is missing.

Next lesson

You have finished the Data Analysis course. Pick another skill to learn next.

Questions

Where do I go next?

Practise on data you care about, and consider the SQL and Excel courses for working with larger tables, or a course on a programming language to go deeper into code.

How do I make my analysis my own?

Choose a real question, write the plan in a few lines, and keep a log of your choices. The habits in this course work on any data, big or small.

What if my data is much bigger than this?

The steps are the same. Bigger data needs more care with the pipeline, tests and checks, and sometimes tools built for scale, but the thinking does not change.

Progress is saved in this browser.