Bias: selection, measurement, confounding (related to both cause and effect) Validity = accuracy, truthfulness: content: includes all dimensions of construct and nothing more, construct: related coherently to other physically unverifiable measures of same phenomena, criterion: can predict directly observable phenomena (e. g. wound). Regression to mean: when see abnormal result, repeat the test, new result is often closer to normal. Criteria for abnormality: statistically unusual, clinically sick, and treatable. To increase prev: look for symptoms, get referrals, search demography. Discrimination ability: plr/nlr, if >50, then good test. Nomogram: draw straight line from pretest prob to lr to post-test prob. Denominator population: should be specific (all at risk, relevant to question, scope well-described, good sampling method) Prevalence: people already sick, just count sick ones, divide by all at risk. Either point (e. g. jan 31: or period (e. g. 1994-1995). Prevalence advantage: used for diagnostic tests, allocation of resource, identify future hypotheses. Prevalence disadvantage: don"t know cause and effect, which precedes which.