PSYC4050 L.7 Logistic Regression

what does the b coefficient tell us in logistic regression?

how the log odds change in response to a unit increase in the predictor variable by exponentiating b

why should the 95% confidence interval contain 1 when evaluating the odds?

if exp b = 1, changing the predictor value produces no change in odds

what are the 2 main questions when evaluating logistic regression models? (same as linear)

is predictor useful? is model good fit?

what does the sequential method entail in logistic regression?

(like hierarchical) predictors entered in blocks, shows predictive value of varibs. in context of predictive value of model

what does the direct enter method entail in logistic regression?

(like linear) all predictions entered simultaneously, evaluates strength of diff. predictors, not for testing specific hyps.

how do you evaluate if the model provides a good fit in logistic reg.? nagelkerke and cox& snell

R2, report range of variance explained by model from both tests

in logistic regression output, what does the -2 log likelihood column mean?

amount of unexplained variance at each iteration (decreases until spss terminates it)

in sequential method in logistic regression, what does the first predictor explain?

as much variance as poss, then others are added to improve fit

in linear regression, what does c mean?

baseline level

how do you interpret equal odds?

both outcomes are equally probable

in logistic regression output, where do you find preliminary info?

case processing summary, dependent variable encoding

how do you evaluate odds (the exp b coefficients) to see if the predictor in logistic regression does anything useful?

check that the 95% confidence interval includes 1, means 95% sure the interval around expb contains the true population parameter for the change in odds

how do you evaluate R2 using homer & lemeshow test? what does it entail?

chi square, how closely model predicts outcome categories, tests against h0 that predictions are perfect

how do you evaluate R2 using omnibus test? what does it entail?

chi square, tests whether all predictors combined account for variance, test against h0 that R2 = 0

how does logistic regression differ from linear?

deals with nonlinear relationships, larger samples are needed, categorical outcome

what are the 3 main methods of logistic regresison?

direct enter, sequential, stepwise

what is a wild test?

form of chi square, sig. result allows rejection of h0

what are the 2 methods of conducting stepwise logistic regression?

forwards (start with 0 predictors, add 1x1), backwards (start with full)

what are the 2 main questions in regression?

good predictor? good model fit?

how do you assess whether the predictor variable does anything useful in logistic regression ?

log odds (b), odds (exp b coefficients), no change in odds

what is the change in odds expressed as?

multiplier (or proportionate change) on old odds

what result do you want from the homer & lemeshow test? what does it mean?

non-sig. = no discrepancies between model & data

what happens to odds if exp(b) is = 1?

odds are constant

what happens to odds if exp(b) is <1?

odds decrease

what happens to odds if exp(b) is >1?

odds increase

in logistic regression output, what results do you report from block 0?

percentage predicted in classification table, which predictors were/werent included

in linear regression, what does y' mean?

predicted value

what does the stepwise method entail in logistic regression?

predictors added sequentially to explore hypotheses, compare model fit after each step via nested model comparisons

what does logistic regression do?

predicts outcome variable on the basis of set of predictor variables, only categorical data

what should you report for logistic regression?

preliminaries, block 0 output, block 1 + output

what are odds?

probability of event/ 1 - probability of event

what should logistic regression equation output?

probability of specific outcome with range from 0-1

in linear reg., how do you test if the model provides a good fit?

r2, test if model accounts for sufficient proportion of variance

in logistic regression output, what are iterations?

series of approximations to arrive at best solution, each step produces more accurate approx.

in logistic regression output, what does the wald statistic tell us?

strength of each predictor & significance compared to 0 magnitude (h0)

how do you compare odds?

take their ratio & divide them

how do you interpret odds greater than 1?

target outcome is less probable than alt.

how do you interpret odds less than 1?

target outcome is more probable than alt.

how does logistic regression use odds?

tells us how the odds of belonging to a certain category change with unit increase in a predictor (b coefficient)

in linear reg., how do you test if the predictor gives us a good understanding of the data?

test is coefficients (bs) are different from 0

what is exp(b)?

the no. we need to multiply the original odds to get the new odds

in linear regression, what does b mean?

weight of predictor variable, regression coefficient (slope of the line)

what are you trying to predict in logistic regression?

which category people belong to given the predictors