Flashcards

lecture 8 logistic regression

Toyin O English 25 cards

Cards in this deck (25)

1If you wanted to predict life satisfaction from job satisfaction, income and leisure spend, what analysis should you carry out?multiple regression
2You want to determine the strength and direction of the relationship between number of siblings and self-confidence what analysis should you carry out?Spearman's correlation or pearson's correlation depending on the distribution of the data on the two variables
3you need a p-value to know if something is...significant
4r squared adjusted...can tell us the strength of a relationship but not the direction.
Tell us us how variability in the population is accounted for by the relationship between the variables
5logistic regressionwhich group are you likely to belong to
6what can we do?- logistic regression looks at the population of something, rather than ctually measuring anything
- looks at likelihood
- hotto turn the two rows of data into a curve? The way we do this, is by working out the proportion for the number of offender an reoffendersa long time
7probabilitychance of getting significant finding by chance
8what do we use continuous predictors for?use continuous predictors to determine the likelihood of someone belonging to a particular group/ category
e.g do extraversion and boredom proneness predict whether someone is likely to be a risk taker or not?
9binary logistic regressionA binomial logistic regression (often referred to simply as logistic regression), predicts the probability that an observation falls into one of two categories of a dichotomous dependent variable based on one or more independent variables that can be either continuous or categorical. If, on the other hand, your dependent variable is a count, see our Poisson regression guide. Alternatively, if you have more than two categories of the dependent variable, see our multinomial logistic regression guide.
10Dv has to becategorical
11IV has to becontinous or categorical
12exampleFor example, you could use binomial logistic regression to understand whether exam performance can be predicted based on revision time, test anxiety and lecture attendance (i.e., where the dependent variable is "exam performance", measured on a dichotomous scale – "passed" or "failed" – and you have three independent variables: "revision time", "test anxiety" and "lecture attendance").
13How much data do you need for logistic regression?You need a lot of data for logistic regression or it doesn't work very well
14—100% of the 69 cases have been processed
No missing data
15—This table tell you how the Dv has been coded - important for interpretation
However you code the DV - SPSS might chnage this - if you didnt use 0 and 1 spss would have to change it
16—This section reports the most basic attempt to predict outcome –
in which all cases are predicted to result in the most common outcome
You do not need to report or specifically study this table
17—This table reports the results of this simple to prediction:
As most prisoners are reconvicted in our sample, the predicted outcome is set to ‘reoffend’.
This results in an accurate prediction in 57% of cases.
-tells us how good our first model is
- tells us what we've observed - shows us that 30 people did not reoffence and 39 did offend
- wht we have predicted is that the most common outcome is that people did reoffend
- so far those who did not offend again we got 0% correct
18—These tables tell us that so far no predictors have been entered into the equation.
don't need to report these particularly
19—This block reports the results of the logistic regression analysis.
Should be better than Block 0.
Displays process of approximation in constructing the model
-2 log likelihood = success of model
Higher value  poorer model
Each iteration:
-2 log likelihood falls
benefit decreases (see 5 & 6)
Coefficients of predictor variables adjusted at each iteration
Block 1 has a lot of info in it
You get iteration history - a summary of how our odel is trying to approximate the data
we want this -2 log likelihood to be as small as possible
20—need to see if our model is significant
Tests general fit of model. Include this in a report:
X² = 56.019, df = 1, p < .001
21—Approximates R²
-2 log likelihood is not usually reported
Cox & Snell R square and Negelkerke R sqaure - these two measures of variance are usually reported - explaining between 57% and 75% variance in the data
22—Summarises the prediction results and should be compared to equivalent in Block 0.
predicted on the basis of our new model that 25 people will be single offenders but actually it was 30 people (25+5)
predicted 34 would be reoffenders but actually it was 39
The model is now a lot better overall - 86% variance explained, compared to 57% - much better at prediciting single offenders, a bit worse for repeat
23—Contains some of the most critical information.
B = coefficients for each predictor variable in the model.
WALD statistic & Sig = usefulness of each predictor variable.
Exp(B) = change in predicted odds for each unit change in the predictor variable.
24—chart worth looking at
visual representation of prediction process
Not usually reported
They should match whats below?
n's shoild be in one place, y's on the other side
- check model is effective
25limitations of logistic regressionThe relationship between the probabilities on the dependent variable is assumed to be sigmoidal (S-shaped).
The method is very sensitive to outliers.
The ratio of sample size to variables needs to be quite high.
Computer packages (esp SPSS) assume that you are describing relationships not making predictions.
 no equivalent of R² Adjusted is given.
Classification tables show results for the
sample only.