BBC’s reality check hosted an article on “Why is unemployment down but benefit claims up?”
They point out something that everyone should know about statistics, the margin of error.
Simply put: if “…there were about 26,000 fewer people unemployed… That figure is based on a survey… the figure for the change in unemployment was right to within plus or minus 77,000. As 26,000 is less than 77,000, we say that the fall in unemployment is not statistically significant.”
In other words, they could have simply made up a number that is below or near 77,000 and it would still technically be insignificant because of the margin of error. In fact you could have not run the survey and it would have given you the same amount of information. In my opinion it is similar to measuring the height of a house with a margin of error of 2-3 inches. If I say my house is 40 feet 3 inches or 40 feet or even 39 feet 11 inches, it is statistically equivalent. It’s about 40 feet high.
It would be ignorant of me to tell you each month that my house has grown by 3 inches since last month, or shrunk two inches, because of XYZ assumption, because it is within the margin of error and therefore “not statistically significant”. One can simple and more accurately say the height has remained statistically the same, or that the change in unemployment has statistically remained the same.
Rather, news headlines will read “XYZ economy added 35,300 jobs last month.” (Common Canadian Headline)
For me, often fluctuations of 35K jobs in an economy with 35 million people (with approximately 20 million eligible for employment) is a 0.1 % fluctuation which is well below the margin of error. A fluctuation of 350,000 jobs is a 1% change in a 35 million person economy. If we are talking about the USA, with about 200 million available workers, 0.1% fluctuation is 200,000. So reporting anything around 200,000 is “not statistically significant”.