Working with Data — Module Assessment

Question 1

According to the workflow, what should happen before you choose data or methods?

Accepted Answer

Frame a specific question that could come back wrong

Answer

Build a model and see what it says

Answer

Clean whatever data is closest to hand

Answer

Design the final chart you want to show

Question 2

Industry surveys such as the 2016 CrowdFlower report and a 2014 New York Times piece suggest that data professionals spend most of their time on what?

Accepted Answer

Finding and preparing data rather than modelling

Answer

Building and tuning statistical models

Answer

Writing up conclusions for stakeholders

Answer

Designing dashboards and visualisations

Question 3

Why can dropping every row that has a missing value distort your analysis?

Accepted Answer

If the data is not missing at random, the remaining rows are biased

Answer

It always leaves too few rows to compute an average

Answer

Deleting rows is technically impossible in most software

Answer

Missing values never affect the final result either way

Question 4

In the 2013 critique of Reinhart and Rogoff's 2010 paper, what did Thomas Herndon and colleagues find?

Accepted Answer

An Excel formula had left several countries out of a key average

Answer

The authors had invented their entire dataset

Answer

The paper used no spreadsheet at all

Answer

The results were confirmed exactly as published

Question 5

What is the central lesson of Anscombe's quartet?

Accepted Answer

Datasets with identical summary statistics can have completely different shapes, so you must plot them.

Answer

Larger datasets always produce more reliable averages.

Answer

Correlation coefficients are the most trustworthy summary of a relationship.

Answer

Outliers should always be removed before analysis.

Question 6

A dataset of website visit times has an average of four minutes, but almost no visitor stays near four minutes. What is the most likely explanation an EDA plot would reveal?

Accepted Answer

The times form two clumps whose midpoint happens to be four minutes.

Answer

The data was entered incorrectly and should be discarded.

Answer

The average is simply wrong and cannot be computed.

Answer

Four minutes is the most common single value in the data.

Question 7

Why is random assignment central to a fair A/B test?

Accepted Answer

It balances known and unknown differences between the groups so any gap reflects your change.

Answer

It guarantees the two groups will have exactly equal outcomes.

Answer

It removes the need to measure any outcome at all.

Answer

It lets you use a much smaller sample than otherwise.

Question 8

An experiment on ten million users finds a new button color raises clicks by 0.001 percent, and the result is highly statistically significant. What is the right takeaway?

Accepted Answer

The difference is real but so small it may not be worth acting on.

Answer

The result must be a mistake because the effect is tiny.

Answer

Statistical significance proves the change is important.

Answer

The sample was too large for the test to be valid.

Question 9

A model predicts last year's sales almost perfectly but does poorly on this year's. What has most likely happened?

Accepted Answer

It has overfit, memorising noise in the training data

Answer

It has underfit the data

Answer

The training data set was too large

Answer

It needs a simpler scoring metric

Question 10

Why did the Netflix Prize score final results on a set of ratings the teams never saw?

Accepted Answer

To stop teams tuning to the scoring data and to measure true prediction on unseen cases

Answer

To make the competition finish faster

Answer

Because the training data was incomplete

Answer

To reduce the size of the prize money

Question 11

A bar chart makes a change from 10 to 9 look like a near-halving. What design choice most likely caused this?

Accepted Answer

The axis was truncated so it did not start at zero

Answer

Too many colours were used

Answer

A line was used instead of bars

Answer

The underlying data were inaccurate

Question 12

What did Edward Tufte mean by 'chartjunk'?

Accepted Answer

Decorative elements that add no information and distract from the data

Answer

Data that has been rounded off

Answer

Charts drawn by hand rather than by computer

Answer

Charts that contain too few data points

Question 13

According to this lesson, what best captures the core of being data-literate?

Accepted Answer

Habitually asking where data came from and what it leaves out

Answer

Mastering advanced statistical formulas

Answer

Refusing to trust any claim supported by numbers

Answer

Being able to build charts and dashboards

Question 14

A promotional report claims a program "improved outcomes by 30 percent" with no baseline, sample size, or margin of error given. What is the most data-literate response?

Accepted Answer

Ask about provenance, comparison, and uncertainty before judging it

Answer

Accept it, since a specific percentage signals rigor

Answer

Dismiss it outright because statistics can say anything

Answer

Recalculate the percentage to check the arithmetic