Chop Suey

In my role as a Statistical Ambassador, I write about statistics.

This article collates short comments about various topics. This includes Germany’s reproduction number, YouGov surveys, representative samples, and more.

Germany’s Reproduction Number

On 22nd June, Sky News reported:

Coronavirus: Germany’s R number rockets again — from 1.79 to 2.88

Germany has a low number of infected people. The estimated reproduction number is sensitive to localised outbreaks. It reflects what has happened 8 to 13 days ago. The Sky News article does not express that the number is an estimate, nor any uncertainty. Also, it is a dimensionless number, and not a “rate”.

This is from the Robert Koch Institute report (which the Sky News article does not link to):

This value reacts sensitively to short-term changes in case numbers, such as those caused by individual outbreaks. This can lead to relatively large fluctuations, especially if the total number of new cases is small. The current estimate of the 4-day R-value is 2.88 (95%-prediction interval: 2.16–3.73) and is based on electronically notified cases as of 21/06/2020, 12:00 AM.

Estimates of the reproduction numbers (R-value and 7-day R value) were between 2 and 3 during the last few days. This is mainly related to local outbreaks which are described above, the outbreak in North Rhine-Westphalia playing a particularly important role in this context. Since case numbers in Germany are generally low, these local outbreaks have a relatively strong influence on the value of the reproduction number.

After those localised outbreaks, Germany’s reproduction number subsided. In their report on 24th June, the RKI estimate of the R-value is 0.56–0.91.

Given the uncertainty, we should round these figures, such as: 0.5–0.9. Their central estimate was 0.7.

A more sober report would have recognised the uncertainty. (Image: Robert Koch Institute)

The “chlorinated chicken pie” post

The pro-EU page ‘Scientists for EU’ shared the following image:

Why is YouGov in all lower-case? (Image: Facebook)

This is misreporting. The question is not whether someone would like to eat ‘chlorinated chicken pie’. The question concerns a future UK-US trade deal. It is whether allowing imports of “chloride washed chicken” would be acceptable. Also, 3D pie charts are not a good way to visualise data.

YouGov asked 1,646 GB adults on 10–11th June 2020. This YouGov survey was via their internet panel:

Why didn’t they just copy this graph? (Image: YouGov)

Do 4 in 10 ‘very liberal’ Americans support changing the flag?

Prof Eric Kaufmann (Birbeck College) claimed that 4 in 10 ‘very liberal’ Americans back changing the national flag. This claim should be treated with great caution.

Prof Kaufmann fielded questions on Amazon MTurk and Prolific:

The sample is not representative of the American population — I used the Amazon Mechanical Turk and Prolific Academic survey platforms that thousands of academics use. Respondents on these platforms lean young, liberal, and white. But as this is precisely the group I wished to study, this is not a major limitation.

This is a strange assertion. ‘Liberal’ or ‘very liberal’ Americans are an attitudinal segment. Imagine asking British people if they are left-wing or right-wing. Those segments will change over time, as people change their views.

Researchers look at these segments as part of a larger, representative survey. It is unclear why ‘very liberal’ people on Amazon MTurk would be like others elsewhere.

The survey is a series of agree-disagree statements, which risks acquiescence bias and expressive responding. Some question items are long, vague and difficult to interpret, such as:

Gradually replace many older public buildings with new structures that don’t perpetuate a Eurocentric order, until a more representative public space is achieved

Once we include those people, 29% of the ‘very liberal’ sub-sample clicked they agreed with:

Move, after public consultation, to a new American flag that better reflects our diversity as a people

An index of cultural change would be interesting to build and study.

Positive Proportion

Nadine Dorries MP (Conservative, Mid Bedfordshire), the Health Minister, wrote:

If 100 people anywhere else in the UK were to be tested, 2 would test positive. In Leicester, that figure is 10.

This is wrong. A small change in words can make a large change in meaning.

These figures appear to be referring to the positive share among those tested.

This is from the PHE report on Leicester. (Image: PHE)

This is different to how many people would test positive in the whole population. Between 14th and 27th June, the ONS estimates 0.04% (0.02% — 0.08%) would test positive for SARS-CoV-2. That survey is for English private households.

Express headlines

In a Daily Express article, the headline was:

Boris Johnson is least popular leader in the WORLD according to poll

This headline is inaccurate. The question is not about each leader’s popularity. It refers to each government’s handling of the coronavirus crisis:

How well or badly do you think the [national government] are handling the issue of COVID-19?

The central estimate was the UK government had the joint-lowest score of the 22 countries. (Image: YouGov)

YouGov survey adults in 22 countries on this topic, not “the WORLD”. Surveys provide estimates, subject to many sources of potential error.

Following contact with IPSO, Express journalists amended this headline.

Do ‘49% of English voters support independence’?

No. Business for Scotland made this claim. The National repeated their error.

The question has an agree-disagree format, of a double-barrelled statement. There are risks of acquiescence bias. The statement was:

England should be an independent country and Scotland, Wales and Northern Ireland should be allowed to stand on their own two feet.

34% agreed with the statement. (Image: Panelbase)

30% of English respondents neither agreed nor disagreed with the statement. The headline excludes these answers. The National adds spurious precision:

The poll revealed 52.5% of Conservative voters in England support independence for the country.

Panelbase surveyed 1,015 English adults on behalf of Business for Scotland. The company collected internet panel responses between 30th June and 3rd July.

Change, change, change

A ‘March for Change’ Facebook post asserted:

Piers was right.

First, those statistics are by date of report, not date of occurrence. Given the time lag between infection and reported death, it is too soon to judge if Morgan was “right”.

Second, the seven-day rolling average (by date of report) has fallen:

  • 5th July: 96
  • 12th July: 86
  • 19th July: 69
  • 26th July: 66

There is a clear weekly cycle in the reporting. We should avoid drawing conclusions from partial weeks.

Third, comparisons between nations and regions need care. There are demographic differences, including in age profile, population size and density. Definitions differ between the four parts of the UK.

Accuracy matters. Share statistics, not misinformation.

This blog looks at the use of statistics in Britain and beyond. It is written by RSS Statistical Ambassador and Chartered Statistician @anthonybmasters.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store