Back to Schedule

Slide 1

LIS 570

Bivariate analysis (continued)

Slide 2

Summary

Prediction

• regression analysis

• comparing means

Statistical inference

• significance

Slide 3

Regression

Slide 4

Regression

Slide 5

Regression equation

Slide 6

Regression

• Prediction using regression is most secure when the independent variable x takes a value within the range of the x values in your data

• not about cause and effect

extrapolation

• Using the regression equation for prediction outside the range of the original data

• less secure

Slide 7

R 2:  the Coefficient of Determination

• Link between correlation and regression

• tells us the proportion of the variance of one variable that can be explained by straight line dependence on the other variable

• How much can we rely on the regression estimates

Slide 8

R 2:  the Coefficient of Determination

• .892 = .79

• 79% of the variance in first year uni marks can be accounted for by the variance in the sample’s SAT scores

• 21% of the variance in first year marks is accounted for by other unknown variables

Eg 2. the correlation between length of car and mpg/l is -.7

• Interpret in terms of r2

• percent of variance in the Y scores variable  which is associated with the variance in the  X scores.

Slide 9

Regression

• Use when...

1.      both the variables are interval

2.      for prediction about the scores of individual cases or groups

3.      to measure the amount of impact or change that one variable produces in another

Slide 10

Comparison of means

• Focus on comparison of data distributions

Slide 11

Comparison of means

• Appropriate when..

	Dependent variable is interval
	independent variable has few categories (2 or 3)
	initial analysis

• look for patterns then use tables

Slide 12

Statistical significance

• “Real” or “Chance”?

• Significance

• judgements that are made according to agreed on mathematical rules of probability

• used to infer observed differences or relationships in the sample to the population studied

Slide 13

Statistical significance

• If we drew 100 samples, how likely is it that we would get a faulty one

• Probability theory

• provides us an estimate of how likely it is that sampling error is the real explanation for the association that we are observing

Tests of significance

• a figure from 0.000 to 1.000

• the probability of error

Slide 14

P - value

• P = 0.04

• in only 4 out of every 100 samples would we expect to see the association we have noted purely by chance.

• The much stronger likelihood is that the association is real

Slide 15

Statistical significance

• Every finding derived from a sample is associated with some probability of error

• How much probability of error should be tolerated?

• Researcher decides

• sometimes referred to as tolerance limits

• 0.05 common

Slide 16

Presenting data

Slide 17

Another example

Slide 18

Means and proportions

• Two means

• T-test

Several means

• Analysis of variance (ANOVA)

Proportions

• Chi-square

Slide 19

Conclusion

• Univariate analysis

• Describing frequency distributions

• shape; central tendency; dispersion

Inferential statistic

• Interval estimates

Bivariate analysis

• cross tabulation; correlation (strength, direction, nature)

• scattergram; regression (prediction)

• statistical significance

• comparison of means (T; ANOVA) and proportions Chi-square