IMC

Any IV with many values. 

You measure or count its level in each case

ISP

An IV with few values. 

You partition your sample into groups, assigning cases according to their score on the variable.

IXB

An IV with few values. 

You  create a few levels of this variable; then expose each case to one level (all cases at the same level are called a group or collection).

IXW

An IV with few values.

You create the levels of this variable; then expse each case to every level of the IXB variable.

DMC

A DV with many values. 

You measure or count its level in each case.

DSP

A DV with few values. 

You partition your sample into groups, assiging cases according to their scores on the variable.

MMC

A mediating variable with many values.

You measure or count its level in each case.

MSP

A mediating variable with few values. 

You parition your sample into groups, assiging cases according to their scores on the variable.

PMC

A precausal variable with many values.

You measure or count its level in each case.

PSP

A precausal variable with few values. 

You partition your sample into groups, assiging cases according to theri scores on the variable.

Types of Surpressed Variables

1. SRA
2. SPC
3. SCA
4. SSC
5. SRM
6. SCO

Surpressed Variable

A variable that does not vary at all, or whose average value is equal across groups or conditions.

Because it is surpressed, this variable cannot play a role in any observed covariation among other variables in the study.

W

Wild Card

A variable that is neither measure nor surpressed.  Thus you can neither study its role in affecting covariation among other variables, nor eliminate such influence.

Factorial Design

An empirical study has a factorial design if...

1. It includes at least 2 categorical IVs
2. One condition within the study presents every possible combination of levels (one level from each IV)
3. It includes at least one DMC or DSP variable

Research Design

logical arrangement of variables in a study. every study has a research design.

strong designs allow relations among variables of interest to be assessed unambiguously and relations not of interest to be isolated or removed.

3 Steps to designing new research

1. Identify all variables
2. Decide each variable's logical status and measurement status.  logival and measurement status indicate treatment mode.
3. arrange the study's variables within a specific research design

Logical Status

a variable's role in any causal interpretation the researcher intends to make of research results.

6 Statuses: I, D, M, P, S, W

Additive Treatments

allow us to learn about how a variable covaries with other variables; how it functions as an aspect of psychology

Subtractive Treatments

Suppress (neutralize/remove) the effects of variables in a study; as a result we cannot learn about such variables, but we can learn more about the other variables.

EXPERIMENTS

• IXB + (random assignment or counter-balanced assignment) + (DMC or DSP)
• IXW + (DMC or DSP)

A Correlation Matrix

A square matrix of statistics from a multivariate study.

Each variable occupies one row and one column of the matrix.

Each cell shows the value of a correlation coefficient between its tow and column variables.

1. Upper Right Triangle
2. Lower Left Triangle
3. Main Diagonal

Upper-Right Triangle

Shows the values of r_xy for each pair of variables

 r_xy =Correlation coefficient

# of Correlations

#Corrs=(#vars)(#vars-1)

          2

Main Diagonal

The Main Diagonal can either...  

• shows r= 1.00 (because any variable can predict its own scores with perfect accuracy)
• cells are left blank b/c redundant to put 1.00
• shows the reliability coefficient for each variable if known (will be strong positives) -- ie. measured variables twice and is degreee of test-retest

Common

Variations

of a correlation matrix

• upper-right triangle gives values of r_xy; lower-left triangle gives N and the significance level for each value of r
• upper-right and lower-left triangles display r values found in different groups or in different samples

Hazard of Multivariable Studies

-May draw results that are correct for your sample but may not be otherwise (increases with more variables)

Groups: Sample divided into groups

Sample-> voting, consensus, individual

Conditions: Whole sample goes through each condition
Sample-> voting -> consensus -> Individual

Suppressed Variable

• Doesn't Vary (remains at fixed level) age
• Varies at a fixed level (average level is same in different groups or conditions of the study)
• Varies but doesn't influence other variables

Pre-Causal

Variable

A variable that may covary with other variable(s); if it does, you lack sufficient evidence to identify either variables as a cause or effect of the other.

Why Supress Variables?

• is as important as manipulation and measurement.
• a subtractive process that strengthens the interpretability of other relationships.

MC

A Measured/Counted Variable

measures how much or count how many of the focal attribute(s) is/are present in each observation; a scalar variable

SP

A Sample Partition Variable

assign each case in the sample to one category of this variable.  This partitions the sample into groups; each group includes all cases with the same attribute on the variable.

Converting:

MC -> SP

Scores on a measured/counted variable can be used to divide a sample into groups. 

First transform scores on the MC variable into just a few (highly rounded) values; then assign each case to a group based on its (highly rounded) score

Dichotomizing a Variable

common way to convert an MC variable into an SP variable.

Establish a cut-point value; call cases who score above it 'high' and those who and those who score below it 'lows'

MEDIAN

SPLIT

most common way to dichotomize a variable. 

Process: Find the median then assign each case to a 'high' or 'low' group depending on whether it scored above ot below the median.

XB

experimentally manipulated 'between groups' variable

researcher creates a categorical variable.  scores are merely labels for the levels of a treatment that was manipulated by the experimenter.

Logical status of XB variables is always "I". Each case is exposed to one level of the IXB variable; then its status is assessed on a DMC or DSP variable

XB example

bake 3 batches of brownies: one usuing butter, one using vegetable shortening; one using fish oil.  Different groups of 10 people taste and then rate each batch.

bake 3 batches of brownies: one usuing butter, one using vegetable shortening; one using fish oil.  Different groups of 10 people taste and then rate each batch.

XB example

XW

experimentally manipulated "within subjects" variable

Same as XB except each case is exposed to all levels of the IXW variable, the DV is measured after each exposure

XW example

bake 3 batches of brownies: one using butter, one using vegetable shortening, one using fish oil.  All 30 subjects taste and rate all three batches. (Probably should have people drink some water between tastings)

bake 3 batches of brownies: one using butter, one using vegetable shortening, one using fish oil.  All 30 subjects taste and rate all three batches. (Probably should have people drink some water between tastings)

XW example

how scores are obtained on a variable. 

6 statuses: MC, SP, XB, XW, S, W

I

Independent variable (I)

a variable is an independent variable (IV) if the researcher believes it controls- at least in part- the level of another measured variable in the study

D

Dependent variable (D)

a dv is being studied in order to determine whether its level is controlled (at least in part) by another measured variable in a study

M

mediating variable (M)

any variable that "carries" the influence of a cause to its effect.  Mediators are postioned in the middle of a causal sequence or chain; in A -> E -> B, E is a mediating variable

NO. Whether or not an IV is a cuase depends on the study's results (and on the soundness of the study's research design).  IV means, in effect, "possible cause".

The Logical Statuses are...

1. I
2. D
3. M
4. P

The Logical-Measurement Statuses are...

1. S
2. W

The Measurement Statuses are...

1. MC
2. SP
3. XB
4. XW

variables you do not want to learn about in a study, but should deal with anyway because they may 'matter' in your topic area

ie: 3rd variables, confounded manipulations

ensure all cases in our sample score the same on any nuisance variables

ex.s

1. all words have 5 letters in word memory task
2. same written text explanations to each partic.
3. every partic. gets same effective dose of alcohol and is test exactly 30mins. later

in building sample, select only cases with particular score (or cases that fall within a limited range on that variable).

Result: sample will not vary on that variable (or will vary on it less than does the population)

Why Suppress?

suppressing nuisance variables clarifies relationships among other variables in the study

When to Suppress?

1. as you select sample (SSS)
2. as you assign cases to groups/conditions (SRA, SCA)
3. during the study's procedures (SPC, SRM, SCO)
4. when you analyze the data (SCC)

SSS

suppression by selective sampling

SRA

suppression by random assignment of cases to conditions

SCA

suppression by counterbalanced assignment of cases to conditions

SPC

suppression by procedural controls

SSC

suppression by statistical controls

SRM

suppression by repeated measures

SCO

suppression by counterbalanced ordering of stimuli

bivariate correlational design

#coefficients=(n)(n-1)

                    2

Intact Groups Comparison Design

an existing ISP variable is used to assign cases to groups which are then compared on other (DMC or DSP) variables

ex.

men vs. women

Putnam's class vs. Templeton's class

Sort-Of Intact Groups Comparison Design

researcher rounds scores on an IMC variable to just a few values which converts it to an ISP variable.  He partitions the sample into groups based on scores on that variable, and compares them on other (DSP or DMC) variables

All of bowdoin into groups by gpa (0-1, 1-2, 2-3, 3-4) and compares them on dependent variables

Cross-Classification Design

Any research that uses a cross-classification table.  Place each case in one cell of each table, look for evidence of covaritation between the variables.

basically cross-tab tables

Coefficient Comparison Design types

2+ DMC variables. H_o sets the value of a correlation coefficient at something other than r_ab =.00

1. r_ab in different groups; study whether two variables are correlated to the same degree in different groups
2. r_ab compared with r_ac; does A correlate equally with two other variables (B and C)? Null hypothesis says it does

1. partial correlation anaylysis
2. mediator analysis
3. moderator analysis
4. multiple correlation
5. factor analysis

a family of research designs with three defining features...

1. cases are assigned to groups randomly (or with counterbalancing)
2. researcher manipulates at least one IV and exposes each case to one level of that IV
3. at least one DMC or DSP variable is assessed after cases have experience the IV

simplest experiment design with one IXB2 variable and one DMC or DSP variable

• randomly assign cases to either the treatment group or the control group
• expose cases in the treatment group only to the experimental stimulus
• measure each case on the DV
• if the groups score differently on the DV, conclude that the IV caused scores on the DV

Ways to ensure IV's for groups ONLY differ on the DV in between-groups designs...

• dummy/filler task: a hollow task for the control group to perform; similar to the treatment but lacking the true IV.  alcohol
• sham surgery: surgery performed on control-group animals; duplicates surgery on treatment animals, but omits the actual 'treatment'

when subjects have expectancies about what will happen in a study, those beliefs can influence thier behavior, independent of any effect of the actual treatment

How to Handle:

• keep subjects unaware of procedure as much as possible.
• compare behavior of people who knowingly experience the treatment with people who falsely believe they've experience the treatment.

PLACEBO

a 'pretend' treatment that leads to the participant to believe (expectancy) he has been given the real treatment

Placebo Designs

Balanced Placebo Design

The Groups:

1. subjects correctly believe they receive tment
2. subjects falsely believe the tmt (actually placebo)
3. subjects correctly believe they are in control grp
4. subjects falsely believe are in control grp (actually get tment)

randomly assign cases to tment groups, then premeasure the DV in all cases.  Manipulate the IV then re-measure the DV.  Compare measurements.

demonstrates that the tment and control groups were equal on the DV before the experiment began.  can also assess whether DV scores change over time, regardless of the effect of the IXB

experiments that compare two ore more treatments to the same control group.

Benefits:

• you get 'double use' from control group
• can directly compare the effects of the two(or more) treatments

Quasi-Experiment

 empirical research that resembles an experiment, but have less control of some aspects of study. provide suggestive evidence of cause-effect relation. useful when cannot perform true experiment.

ie, train experiment

Intact-Groups Quasi-Experiment

groups were not created by random or counter-balanced asignment of cases to conditions. if groups created by IXB perfom differently on DV then researcher cannot be sure whether IXB (or preexisting & unmeasured differences between cases in different treatment groups caused the gorup differences on the DV.

Within-Subjects/ Repeated-Measures Designs

a study in which a dv is measured multiple times in the same cases.

Benefits:

• sample size is N for every level of IXW providing more data than b/ween grps design
• repeated measurement of a DV in the same cases lets you surpress subject variables. SRM

Within-Subjects/ Repeated-Measures Experiments

• study has one group with all the cases
• expose each case to one level of IXW, then measure DV
• expose to another level of IXW; measure
• repeat for rest of IXWs
• compare sample's performance on the different measurements of the DV

Repeated-Measures in a Between-Groups experiment

Order Effects

how the order of conditions effects the study.

solve by switching up the order.

Factor

Cell

for a "3 X 3 X 2" factorial design

1. # of Factors?
2. # of levels for each factor?
3. # of cells in design

1. 3
2. 3, 3, and 2
3. 18

Cell Means

Marginal Mean

the mean DV score among all cases at one level of the factor

(end of the rows)