white printer paper with pen
madeleineshort

What is a 95% confidence interval?

Question: A lot of the study findings you mention include 95% CI.  I’m not sure I understand what that means. Would you please explain?

Answer: Absolutely!  The 95% CI (confidence interval) is a statistical term related to our uncertainty about a given estimate.  In statistics, we’re using a set of data to make inferences about a bigger population.  A point estimate is our best approximation of the truth and the 95% confidence interval is the range of values that we’re quite certain (95% certain) that the true estimate falls within.  Generally, the wider the confidence interval, the less certain we are in the validity of our point estimate; the narrower the confidence interval, the more certain we are in the validity of our point estimate.  It’s important to pay attention to confidence intervals, especially because they are important reminders of the limitations of the point estimates.  I’ll provide an example to elaborate.  The other day, I cited some data on increased risk of hospitalizations for those infected with the B.1.1.7 variant, “Controlling for age, sex, region, comorbidities, and several other factors, researchers found that those who were infected with the B.1.1.7 variant had 1.64x higher odds of hospital admission (95% confidence interval 1.32-2.04).”  Here:

  • Our best approximation of the true value, the point estimate, is 1.64.  That is, people infected with B.1.1.7 are 1.64x more likely to be admitted to the hospital as compared with those infected with a non-variant form of SARS-CoV-2.  But, because we’re using a smaller set of data to make inferences about a larger population, we can’t be totally sure of our point estimate, hence the 95% CI….  
  • We are 95% confident that the true value is between 1.32-2.04.  That is, we are 95% confident that if we had data on everyone infected with B.1.1.7 variant to compare with data on everyone infected with the non-variant form, we would find that at a minimum, people infected with B.1.1.7 are 1.32x more likely to be admitted to the hospital and at a maximum, they are 2.04x more likely to be admitted to the hospital.