Read on Twitter
People are into math and election lawsuits these days, so here's a thread on one of the statistical claims in the infamous #kraken lawsuits being brought by Sidney Powell and her merry band of misfits, the law side of which is well covered by @questauthority and @AkivaMCohen
Specifically, the “expert report” by William Briggs, statistician to the stars (his resume in his affidavit actually says that). This report is based on data from Matt Braynard, who is a Trump operative seeking fraud in the 2020 election.
Braynard collects his data by basically calling voters and asking them if they received a ballot they did not request, or if they sent a mail-in ballot back that was never received.
Briggs looks at the data and says, Braynard talked to 100 people, and 5 people claim they received a ballot they didn’t ask for and 5 claim they sent one back that never got counted.
Based on those proportions, he runs a basic confidence interval to calculate that a very high number of ballots were fraudulent. The statistics he runs are TECHNICALLY perfectly valid and any first year stats student could confirm the same analysis. HOWEVER…
This analysis has an ABSOLUTELY ESSENTIAL REQUIRED assumption that the underlying data is UNBIASED and REPRESENTATIVE of the population. While Braynard’s data is basically a CLINIC in how to collect biased data, an example to illustrate why…
Suppose I am conducting a survey about violent crime, I call 1000 people, and say “Hello, I’m conducting a survey on violent crime, can I ask you some questions?”
900 hang up or refuse, 100 talk to me, out of that 100, 10 say they experienced violent crime. I then run my confidence interval based on that 10 of 100 proportion and loudly declare “I’m 95% certain that between 5% and 17.6% of the population have been a victim of violent crime”
I go on to say, “this is far higher than the roughly 1% the Govt estimates, therefore, based on my technically correct statistics, the govt is covering up violent crime”
What did I miss? Well clearly I missed the 900 people who hung up on me. They had no interest in violent crime, and the people who stayed on were BIASED towards people who had an interest in violent crime because they were victims of it.
Or people who are worried about violent crime and are willing to lie about it over the phone. Or you know, just people who are bored and want attention.
In my analysis I’ve made the clearly flawed and fatal assumption that the population of people who answered vs didn’t answer is roughly the same. This bad assumption ENTIRELY invalidates my analysis. Throw the whole damn thing out.
This is exactly what Braynard did in his data collection and what Briggs did in his analysis. No competent or well intentioned statistician would reach a conclusion based on Braynard’s clearly flawed data.
Running the calculations is easy. Near anyone can do it. Collecting unbiased and representative data is the hard part of statistics that has been completely and intentionally disregarded in this analysis. This is how people use technically correct statistics to lie.
This is why the saying exists, “There’s truth, lies, and statistics”
To pile on a little bit, another “expert” in this case for Powell is a guy named Dr Shiva, who you may know from the fraud “hearings” at the Ramada inn. Know who thinks Dr Shiva is a hack? William Briggs. https://wmbriggs.com/post/33578/
