Medical statistics 2_ chance.pdf

18/02/24 Medical statistics 2: chance Learning objective: explain what random error and systematic error are, and how they affect the precision and validity of epidemiological measures. Learning objective: use and interpret t wo common ways of quantifying random error: p-values and con dence inter v...

18/02/24 Medical statistics 2: chance Learning objective: explain what random error and systematic error are, and how they affect the precision and validity of epidemiological measures. Learning objective: use and interpret t wo common ways of quantifying random error: p-values and con dence inter vals. Learning objective: de ne and distinguish type I error, type II error and power. Learning objective: understand the difference bet ween statistical and clinical signi cance. Background/review: What can be wrong in a study? measure truth Random error 2 swarm systematic Information Bias confoundi Focus of this lecture. Types of error: Random error: When the obser ved value of a measure deviates from its true value by chance. This reduces the precision of the estimate. Precision is how close the values are to each other. Systematic error: When the obser ved value of a measure deviates from its true value in a consistent direction. This reduces the validity of the estimate. Validity is how close the value is to the truth. Random error: Variation in the result of a study that is due to chance alone. It is unavoidable and inherent in any measurements or obser vation. It can be caused by: 1. Sampling error: the difference bet ween the sample and the population. 2. Measurement error: the inaccuracy or imprecision of the instruments or methods used. Imagine we want to estimate the average height of adult women in the population. If we take several different samples of women from the population, the answer may differ from the truth, if by chance, they are shorter than average, our result will be too low, and if by chance they are taller than the average our result will be too high. This is known as sampling variability ( error ). Study size and error: Measuring random chance: Hypothesis testing: The null hypothesis is assuming there is no relation bet ween A and B. The alternative hypothesis assumes there is a relation bet ween A and B. It is easier to disprove a statement than prove it is true. P-values: This is a probability ranging from 0 to 1. Comes from a statistical test testing a particular null hypothesis. 0.05 or 5% is commonly used as a cut-off such that if P