Flashcards in BIO 330 Deck (379)

Loading flashcards...

331

## Fisher's exact test assumptions

### random samples

332

## Fisher's Ho

### state of A and B are independent

333

## conduct Fisher's

###
–list all possible 2x2 tables w/ results as or more extreme than observed table

–p-value is sum of the Pr of all extreme tables under Ho of independence

–assess null

334

## Computer-Intensive methods

###
cheap speed

hypothesis testing- simulation, permutation (randomization)

standard errors, CI- bootstrapping

335

## hypothesis testing, simulation

###
–simulates sampling process many times- generate null distribution from simulated data

–creates a 'population' w/ parameter values specified by Ho

–used commonly when null distr. unknown

336

## simulation to generate null distribution

###
1.create and sample imaginary population w/ parameter values as specified by Ho

2.calculate test statistic on simulated sample

3.repeat 1&2 large number of times

4.gather all simulated test statistic values to form null distr.

5.compare test statistic from data to null distr. to approx. p-value and assess Ho

337

## generated null distribution

###
P-value ~ fraction of simulated X^2 values ≥ observed X^2

none ≥ observed, P < 0.0001

338

## Permutation tests (Randomization test)

### test hypotheses of association between 2 variables; randomization done w/o replacement; needs 'parameter' for association btw 2 variables

339

## Permutation test used when

### assumption of other methods are not met or null distribution is unknown

340

## Permutation steps

###
1.Create permuted data set w/ response variable randomly shuffled w/o replacement

2.calculate measure of association for permuted sample

3. repeat 1&2 large number of times

4. Gather all permuted values of test statistic to form null distribution

5. Determine approximate P-value and assess Ho

341

## Bootstrapping

###
calculate SE or CI for parameter estimate

useful if no formula or if distribution unknown

randomly 'resamples' from the data with replacement to estimate SE or CI

ex. median

342

## bootstrapping steps

###
1.random sample w/ replacement- 1st bootstrap sample

2.calculate estimate using bootstrap sample

3.repeat many times

4.calculate bootstrap SE

*only sampling from original sample values

343

## simulation

### mimics repeated sampling under Ho

344

## permutation

### randomly reassigns observed values for one of two variables

345

## bootstrapping

### used to calculate SE by resampling from the data set

346

## Jack-knifing

### leave-one-out method for calculating SE

347

## Jack-knifing

###
gives same result every time (unlike boot strapping)

calculates mean from n-1, then n-2, then n-3

348

## statistical significance

### observed difference (effect) are not likely due to random chance

349

## practical significance

### is the difference (effect) large enough to be important or of value in a practical sense

350

## effect size

###
ES– degree or strength of effect

ex. magnitude of relationship btw 2 variables

3 ways to quantify

351

## 3 ways to quantify ES

###
standardized mean difference

correlation

odds-ratio

352

## standardized ean difference

### Cohen's d

353

## can find statistical significance

### with a large n, which may not be large effect size, and may not be significant at lower n

354

## Quantifying ES

###
2% difference btw population and sample means

difficult to interpret mean differences w/o accounting for variance (s^2)

Cohen standardized ES w/ variance

355

## Cohen's d

###
simplest measure of ES

difference btw means / Sp

standardizes, puts all results on same scale (makes meta-analysis possible)

356

## Meta-analysis

###
analysis of analysis

synthesis of multiple studies on a topic that gives an overall conclusion; increases sig. of individual studies (larger n)

black line = 1-1 line - no difference, no more, no less

357

## steps in meta-anlysis

### define question to create one large study- general or specific; review literature to collect all studies- exhaustively; compute effect sizes and mean ES across al studies; look for effects of study quality

358

## literature search

### beware of 'garbage in, garbage out', publication bias, file-drawer problem

359

## publication bias

### bias- studies that weren't published- lower n, insignificant, low effect

360