Any statisticians out there?

[Rowanapp]

So I’ve been reading a lot about the death figures on public health England reports. There is lots of stuff about individuals days and weeks where excess deaths were high. I understand excess deaths in 2020 were 14% above the five year average. What I would like to know is is this a statistically significant rise and what is the p value and confidence interval. I know there are some knowledgable people on here and was wondering if anyone knew if this is published anywhere.

A bit late and not quite what you were asking but maybe this will be of interest, @Rowanapp?

www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/bulletins/changesinallcauseandcovid19mortalityovertimeenglandandwales/deathsoccurringbetween28december2019and4september2020#measuring-the-data

Yes, it is statistically significant, both on the overall level and (more frequently) when tested on more granular basis against the historical data.
Asking for a p-value as such is meaningless as mortality / morbidity studies usually involve complex corrections for multiple hypothesis testing, such as for weekly series, so there's no single p-value.

Have you got a link to that?
I’m not asking for the granular detailed levels....
Just 14% above the average yearly no of deaths, seems like it might fall within the expected levels of year on year variation. Maybe that’s because of pandemic control measures, wouldn’t know. But it seems to me the media are reporting it like the death toll is massively above deaths expected in a normal year and I’m not sure that’s true. The ONS site is referring to weekly figures, causes of death, etc etc.
I just want to know, quite simply, were the excess deaths in 2020 more than we could expect with natural variation and with what confidence can we say so. Mumsnet has a lot of knowledgeable people so I thought someone might know where this was published.

Have you seen the Nightingale chart for deaths posted of the Oxford CEBM website:

www.cebm.net/covid-19/covid-19-florence-nightingales-daigrams-for-deaths/

It does not give p values but is a very clear representation of excess deaths since Covid compared to previous years (2015-2021).

I can recommend the Covid 19 Actuaries response group for lots of analysis of death data among other things. Also the continuous mortality investigation bureau monitoring documents.

But spoiler here - yes deaths are statistically significant. Just looking at the graphs presented compared to the last 5 years will show that.

@OneInEight

Have you seen the Nightingale chart for deaths posted of the Oxford CEBM website:

www.cebm.net/covid-19/covid-19-florence-nightingales-daigrams-for-deaths/

It does not give p values but is a very clear representation of excess deaths since Covid compared to previous years (2015-2021).

Why is there in every year a big jump between week 52 and week 1 of the following year?

Why is there in every year a big jump between week 52 and week 1 of the following year?
The data is arranged by the date of reporting, not the actual date of death. There's always a big delay with registrations between Christmas and New Year, all these deaths are usually processed only in the following year.

Thanks, though it'd be something like that.

@Rowanapp

I can recommend CMI Bureau for detailed data on mortality statistics, not sure whether they have published anything specifically on the significance levels in the format you are interested in. Or yes, the Actuaries Covid response group, although they are a bit specialised.

Annual data is not usually looked at much for causal analysis, unless for aggregate retrospective studies / mortality tables construction, as there is significant variation of mortality within the year, winter death rates can be twice the level of summer ones even in a "normal" year. p-values on the annual basis also won't tell you much and might even mislead you, it is a flawed tool to use in this case - there's simply not enough data to ensure good predictive power of such a test, you can't really treat individual years even from the last half century as an independent and identically distributed sample.

Here, a quick and dirty graph from the public data. There's no need for a formal statistical test to clearly identify atypical weeks. What we don't know yet is whether this extra mass of deaths will be offset by a lower overall mortality in the next few years (or in other words, whether people who died during that peak would be the ones who were at the end of their lives in any case). But it is largely a question of curiosity that could be easily answered, I am sure there's an undergrad thesis or two being written somewhere on the topic at the moment.