Puts the data into rank order, but doesn’t tell you anything about the distance between data points.  Nonparametric.

There is equal distance between the data points.  E.g.: Fahrenheit temperature.  There is no meaningful zero.  You can do mathematical calculations, there are relative positions (e.g. if you add a constant to everyone the distribution stays the same).  Parametric.  Fits into a normal curve.

Equal distance between the data points and absolute zero. There is absence of a characteristic (e.g. zero inches of rainfall in a year).  You can do addition, subtraction and multiplication and division.  Parametric.  Fits into a normal curve.

 All measures of central tendency are in the same place. Curves are normal if they are symmetrical.  Equal variability and distribution around the mean

 There is an outlier who has a much higher score than the rest which brings the mean much higher.  It then makes the mean not a good measure of central tendency. 

[image]

 The middle score.  With a skewed distribution, this is the best measure to look at for central tendency.  (e.g. income)

How scores vary in a uniform way above and below the mean

·         68 % of the population is between one standard deviation above and below the mean

·         95 % of the population is between two standard deviations above and below the mean

The central limit theorem is based on a distribution of sample

means.  We will take multiple samples of a population which will

vary around the mean.  Eventually you will get a distribution of

sample means.  As sample size increases, this distribution will

become a normal curve.

Sometimes called the research hypothesis, experimental hypothesis):  States that there is a difference between the treatment conditions. 

You use this test when you are saying there is a difference and you are specifying the direction of the test- e.g. one group will perform better than the other. 

When I reject the null hypothesis but there really is no difference (if I did then measure the whole population).  You reject it when it’s really true.  I say there’s a difference when there is not.  Involves alpha.  Reject when true is not type II. 

 I failed to reject the null when I should have.  I say there is no difference between conditions, but when I measure everyone in the population I did find a difference.  You cannot control for Type II error because you can never tell how many people from the population you will need to sample to find a difference. 

When you are doing a statistical test, you have to be

comparing groups who have similar variability or else it

won’t work (e.g. you can’t compare a leptokurtic vs.

platykurtic). 

The difference between two groups or two distributions

of scores.  It is the appropriate test for testing difference

in means of two populations

This test is appropriate for testing differences in means for several groups to determine if they come from different populations, i.e. whether population means are equal or not.  Just tells you whether there is a difference between any two, but doesn't tell you which.  Have to do a post-hoc test

 Use a MANOVA to control for Type I error.  It is used

with more than one DV (e.g. several scores for each

subject).  Every additional analysis done with an

ANOVA increases error (e.g. 3 ANOVAs at 0.05

= .15 error).  MANOVA combines all three into one

analysis and reduces the error. 

This test is used when a study involves multiple (two or more) independent variables.  There may be a main effect and/or an interaction effect.

This test is used to see if there a significant difference in classifications according to certain categories (e.g. eye color). The Chi Square tells us: Is what we observe significantly different from what we expected? 

This is the square of the correlation and it tells you how good the correlation is.  (R²).  So if correlation is .91 then .91 x .91 = .83 or 83 % shared variability or relationship between variables.

 A special use of correlation coefficients and you develop a model ahead of time and based on correlations throughout the model you are implying a cause and effect relationships:  Two classes- Unidirectional (Path Analysis) and Bidirectional (LISREL) Linear Structural Relation Analyses)

 A type of structural equation modeling.  It's a unidirectional Technique (going from point A to point B).  Based on pairs of variables along the path we are implying there is a cause and effect relationship between variables. 

A technique based on a long term understanding of research in a given area.  Given this knowledge, you make a trend prediction on difference between specific groups, predict the magnitude of these differences, and look at interactions and you will see nonlinear outcomes (e.g. “cubic” and “quadratic”). 

Two baselines, two treatment phases.  Also called a reversal design,  or withdrawal design.  You take baseline data, then implement the treatment then withdrawal the treatment to see if they go back to their baseline data to see if they revert to their original behavior.  Helps to show that your treatment is what really caused the change.  Can be unethical

A measure of how much a person's obtained score is made up of the true score. 

0.90 = 90 % of performance is due to true ability, and 10 % is due to other things.