Statistics - lesson 4 | HackIndia Skills

Key point

One idea

A link in your data does not prove one thing caused the other. It only says two things moved together in your sample.

Use causal words like “caused”, “improved”, or “reduced” only when your design supports it, such as a planned A/B test or a strong comparison.

Link versus cause

A link is a pattern. For example, students who used your timetable app more often also reported fewer missed classes.

A cause is a stronger claim. It says using the app led to fewer missed classes. That needs more evidence because other reasons may explain the pattern.

Common mistake

Common overclaim

Beginners often turn a chart into a story that sounds stronger than the data. “Users who saved goals completed more tasks, so goal saving increases productivity” is too strong if users chose the feature themselves.

Maybe serious users were already more productive. Maybe final-year students used the feature more. Maybe people who had fewer labs had more time to complete tasks.

Your to-do

Check confounders

Do each one yourself, then tap it to tick it off. The ticks are only a checklist for you: they are not marked or scored.

List other factors that could affect the result.
Check if the groups differ on those factors.
Compare within a smaller similar group when possible.
Use careful language if you cannot rule out other explanations.
Save this reasoning in your project README or evidence section.

0 of 5 done

Example

Safer insight

Suppose your hackathon project studies whether reminder notifications are linked with assignment submission. Your data is observational because users turned reminders on by choice.

You can still share a useful insight. Just avoid making it sound like a proven cause.

Data summary:
- 120 student users
- 52 turned reminders on
- 68 left reminders off
- 38% of reminder users submitted before the deadline
- 21% of non-reminder users submitted before the deadline

Too strong:
Reminders increased early submissions by 17 percentage points.

Safer:
In our sample, users who turned reminders on had a higher early-submission rate, 38% versus 21%.
This does not prove reminders caused the difference because users chose whether to enable them.
A possible confounder is study discipline or workload.

Tip

Use AI carefully

You can ask ChatGPT, Gemini, or Claude to help rewrite your insight. Do not ask it to “make the result sound impressive”. Ask it to make the claim honest.

Check the output for banned causal words. If your data was not an experiment, replace “caused”, “led to”, and “improved” with “was linked with”, “was associated with”, or “in our sample, users who…”

Prompt:
Rewrite this project insight so it does not claim causation. Keep numbers. Mention one limitation.

Insight:
Users who enabled reminders submitted earlier, so reminders improved deadline behaviour.

Data:
38% early submission with reminders, 21% without reminders. Users chose whether to enable reminders.

When cause is stronger

You get closer to cause when you control who gets what. For example, randomly show reminders to half of new users and compare early submissions.

That is lesson 5. For now, your job is to label ordinary project data honestly. Say what you observed, name what else could explain it, and avoid overclaiming.