The Office for National Statistics weekly reports feature a graph of death registrations. Readers can see the 2015–2019 range of registrations, plus the latest year. We can emulate this graph for other countries.

Our World in Data collates figures on all-cause mortality from two sources. University of California researchers maintain the Human Mortality database. This database relies on Eurostat and national statistics offices. Ariel Karlinsky and Dmitry Kobak put together the World Mortality data set.

The example country is Sweden. Statistics Sweden publishes weekly updates to its deaths file. Unlike the ONS, the preliminary figures show deaths by date of…

In the New Statesman, George Eaton writes:

On 12 January, one of the grimmest findings of the Covid-19 pandemic was published: “excess deaths” in England and Wales were reported to have risen to their highest level since the Second World War. In 2020, 608,002 deaths were registered, the largest number since the 1918 Spanish flu and 75,925 more deaths than occurred on average over the preceding five years (the technical definition of excess deaths).

Excess deaths are deaths above a baseline. The Office for National Statistics calculate a simple average of the past five years. The ONS continues to use…

All that glitters is not gold. News websites claimed a study suggested Prince William is the “world’s sexiest bald man”. There were news articles on The Sun, Daily Mirror, and indy100 websites.

Who did this study and what did they do?

A cosmetics firm typed names of bald male celebrities and ‘sexy’ into Google. A person counted the number of search results.

Despite this method, some news articles refer to it as a ‘Google study’. This is incorrect. Google neither commissioned nor conducted this study.

What is the problem with this study?

The measure — the number of search results — has no validity. …

Measuring mortality matters. Death is the final outcome for many health problems. There are three different measures which analysts look at:

• Death registrations: this is the total number of deaths. These figures can be by the date of death or registration date.
• Crude mortality rates: the number of deaths divided by a population estimate. This number is often put for every 100,000 people or 1,000 people.
• Age-standardised mortality rates: this calculation starts with mortality rates for each age group. Analysts then calculate a weighted average of all those age-specific rates. That creates the age-standardised mortality rate.

The Office for National…

Surveys provide estimates, subject to many sources of potential error.

How should researchers express uncertainty with survey results? The standard way to show an interval around each central estimate. There are many different kinds of intervals for estimates of proportions.

Confidence intervals

Surveys draw a sample from the wider population. Each sample is one instance of the sampling distribution. Sample statistics can differ from true population values.

A 95% confidence interval has a technical meaning. If you did the survey 100 times, expect about 95 calculated intervals to include the actual value.

It does not mean a single interval has 95% chance…

CODEX

Diagnostic tests are imperfect. In my co-written article in The Observer:

So out of those five positive tests, at least three will be false positives.

This was an illustrative elision, assuming average inaccuracies. Instead, what if inaccuracies varied? What is the distribution of the test’s precision — the true share of all positive results?

Tests don’t have to be testing

Suppose we had the following situation for a binary diagnostic test:

• Prevalence: the proportion of infected people was 0.04%, or 400 in one million.
• False-positive rate: for an average 10,000 tests of uninfected people, three are positive (0.03%).
• False-negative rate: for every test of an infected…

CODEX

Disturbance began after a Scotland on Sunday headline claimed:

Poll suggests No vote lead as Salmond inquiry drives voters away from Scottish independence

Savanta ComRes conducted the online survey of 1,105 Scottish adults (16+), on 4th — 6th March 2021. This survey did not weigh by turnout likelihood:

The voting intention was not weighted by likelihood to vote and should not be treated as a headline Savanta ComRes voting intention.

That introduced in-comparability with previous polls from the same company. What does ‘weighing by turnout likelihood’ mean?

Following the method

Opinion polls seek a representative sample. Companies have a demographic profile of their…

There are several studies of vaccine effectiveness in England and Scotland. There have also been analyses of the vaccine roll-out in Israel.

Despite having similar names, vaccine effectiveness is a different concept to efficacy.

Vaccine efficacy is about reductions in disease. A trial compares vaccinated volunteers to a control group. The trial places people into groups at random. Researchers estimate how well the vaccine does at reducing disease. Trials can under-represent or exclude some people. That affects translation from participants to population.

Vaccine effectiveness is about the programme. How do vaccine programmes affect health outcomes, outside of trial conditions? …