L5 sources of var Flashcards
LO2: disc how obs epidem quants depart from true vals due to rand var. true not necess expected.
-obs val is best estim of true /tendency val but never know what true val act is. rand samps OBS rand var.
LO3: desc how obs vals help tow knowl of true by:
allowing to test hyp about true vals.
-hyp=statem that underly tend of sci int takes a partic quant val. what you think it could be, will never know.
eg preval. new drug neither better/worse than stand tx (ratio surv rates=1). new op neither better/worse post op pain (diff pain scores=0).
-formal hyp testing- how likely is obs if assume hyp true. calc prob of obs as extreme as or more extr than the one obs assuming stated hyp true. if prob small= data and hyp incompat. either someth v unlikely occ and hyp true OR hyp wrong. this prob=p val.
p less than 0.05=rejec hyp at 95% lev. conclus data inconsis with hyp so strng ev ag null hyp. less than 1 in 20 chance occ by chance. sunbstantive ev ag hyp. reasonab to rejec hyp. obs stat sig.
p over 0.05 non of above. NOT prove hyp but fail to rejec.
-lims of testing- stat sig dep on samp size. stat sig not mean clinic imp. stat sig= res of test not likely to have occ by chance alone, more confid that obs a true res.
null hyp RR=1. not prove underly rates really same just that obs diff would be consis with identic underly rates and rand var.
-us practic to test ag null hyp ie ass 2 things are eq or no eff/diff. no prior belief. p less than 0.05 not prove null false but strong sugg.
p over 0.05 not prove true but that obs findings consis with null true.
LO4: obs vals help true val by:
allow calc CI that gives range which incls the true val with specif probab.
-eg service managem. underly tends and what imply about patts of dis and h/c need in gen pop. as well as sys var, ev they obs is aff by rand var.
-estim given var- alm all vals in med subj to chance var. service dem, mort/incid rate, RRs. rational inferences about real size given var. CI estim deg var. best guess at true ratio is mid of CI aka obs val as are best guess.
CI exps precis of estim. range in which true res from pop lies 95% time. narrower= more precise estim. always uncert as samp not whole pop. 95% can draw conclus from samps. comps 2 grps odds ratios, if CI incl 1= no diff btw grps. if absol meas then diff in means incl 0= no diff.
-vals in 95 CI consis with data. if 1 in CI then 1 is poss true val so obs diff from null may be due to chance.
-null hyp val inside CI= p over 0.05.
outiside CI=p under 0.05.
-calc 95 CI. calc obs val eg RR, then err fac.
lower CI= obs val/ef. upper=obs val x ef.
incr cases= more sure about underly val. incr cases NOT just POP.
when calc ef of rate- use numb of cases not rate.
when calc ef RR- use numb cases in both pops.
ef for SMR- use only obs deaths.
-SMR= O/E x 100 us rel to 100.
E- based on stand pop. age sex specif rates applied to study pop age sex grps.
LO1: disting btw OBS epidem quants (incid, preval, IRR), and their true underly vals.
-tend vs obs- coins tend eq but obs may depart due to rand var. eg av cases per mnth but rand var. % complics.
