When you are reading research papers you may come across confidence intervals. For example, one study reported an apparent prevalence of diabetes in UK dogs of 0.34% (95% CI 0.31% to 0.37%) (Mattin et al, 2014). What does this mean? It means the authors estimate that the prevalence of diabetes in UK dogs is between 0.31 and 0.37%. Another study found that the median survival time for animals treated with trilostane was 353 days (95% CI 95-528 days) (Helm et al, 2011). This means that the author’s estimate the median survival time was between 95 and 528 days. These ranges reflect a level of uncertainty around the estimates (the prevalence in the first case and the median survival time in the second case). The uncertainty tends to get smaller as the variability between individuals decreases and as the sample size increases. Importantly though, as Altman and Bland (2014) point out, confidence intervals only reflect the uncertainty due to sampling error. Other forms of uncertainty can arise because of bias.
If you are still unsure about what confidence intervals are, here is a video with good explanations from Dr Nic of the Statistics Learning Centre.
- Mattin et al (2014). An epidemiological study of diabetes mellitus in dogs attending first opinion practice in the UK. Veterinary Record, 174 (14): 349
- Helm et al (2011). A comparison of factors that influence survival in dogs with adrenal-dependent hyperadrenocorticism treated with mitotane or trilostane. Journal Of Veterinary Internal Medicine, 25 (2): 251-260
- Altman and Bland (2014). Uncertainty and sampling error BMJ 2014; 349:g7064