Statistics And Research Design

Front 1

Three Factors that can cause variability in the study's DEPENDENT variable

Back 1

1) the independent variable (experimental variance)

2) systematic error (error due to extraneous variables)

3)random error (error due to random fluctuations in subjects, experimental conditions, methods of measurement etc.)

Front 2

Extraneous (confounding) variables

Back 2

Sources of SYSTEMATIC ERROR

Front 3

Techniques to control the effects of extraneous variables

Back 3

1)radomization (random assignment)
2)Hold the extraneous variable constant (select subjects that are homogenous with respect to the variable)
3)matching
4)blocking (builds the extraneous variable into the study; extraneous variable can then be STATISTICALLY analyzed.
5)Statistical control of the extraneous variable (uses ANCOVA or similar to STATISTICALLY CONTROL, or statistically remove the variability in the DV that is due to the extraneous variable)

Front 4

Minimizing Random error

Back 4

--experimental research, esp. true research design allows the investigator to minimize the effects of random/unpredictable fluctuations in subjects, conditions, and measuring instruments

-Utilizing reliable measuring devices

Front 5

Internal Validity

Back 5

Is there a relationship between the IV and the DV? If so, is the relationship a causal one?

-must control the effects of the IV, control the effects of extraneous variables, and/or minimize the effects of random error

Front 6

Threats to Internal validity

Back 6

1)Maturation (changes within the subject)
2)History (something that occurs/happens/impacts that is external to the subjects)
3)Statistical Regression
4) Selection/Assignment (selection can act alone or interact with other validity threats)
5)Testing
6)Instrumentation
7)Attrition (mortality)

Front 7

External Validity

Back 7

generalizability

Can the relationship between the IV and DV be generalized to other ppl, settings, times etc?

Population validity= generalizability to other people
Ecological validity= gener. to other settings

A study's external validity IS ALWAYS LIMITED BY ITS INTERNAL VALIDITY!!!

BUT a high degree of internal validity does not guarantee external validity.

Front 8

Threats to External Validity

Back 8

1)Interaction between testing and treatment (e.g., pretest can "sensitize" subjects to the purpose of the study)

2)Interaction between selection and tx (e.g., the use of volunteers, they may not reflect the population at large)

3)Reactivity (e.g., ppl respond in a certain way b/c they know thay are being observed; also includes evaluation apprehension, demand characteristics and experimenter expectancy)

4)Multiple Tx Interference (order effects or carryover effects; a prob when subjects are exposed to two or more levels of the IV such as when using a within-subjects design)

Front 9

Between Groups Designs

Back 9

Between groups (or between-subjects) design is used when the effects of different levels of an IV are assessed by administering each level to a DIFFERENT group and then comparing the status or performance of the groups on the DV

Front 10

Within-Subjects Design

Back 10

Repeated measures design

all levels of the IV are administered sequentially to all subjects

Comparisons are made within subjects rather than between groups of subjects

one type is the single-group, time-series design

remember that a single group time series design can help control MATURATION effects but not History effects.

Front 11

Autocorrelation

Back 11

A disadvantage of time series or other within subjects designs

Confounds analysis because the subjects performance on a post test is like correlated with his performance on the pre test; inflates the value of the inferential statistic and makes a TYPE I ERROR MORE LIKELY

Front 12

Mixed Design

Back 12

Utilizes both between groups and within groups methodologies

Front 13

Single subjects design

Back 13

***each single subject design includes at least an A phase (Baseline) and a B phase (treatment)

each subject acts as his/her own no-treatment control

DV is measured througout the study

Front 14

Parametric Tests

Back 14

include t-test and anova

evaluate hypotheses about population means, variance or other parameters

the variable of interest must be measured on an interval or ratio scale

TWO ASSUMPTIONS
1)value of interest is NORMALLY DISTRIBUTED
2)when a study incldues more than one group, there is homoscedastticity (variance of the popuations that the different groups represent are equal)

Front 15

Homoscedasticity

Back 15

assumption that the variance of the populations that two plus groups represent are equal

Front 16

Nonparametric tests

Back 16

used to analyze data collected on variables that have been measured on a nominal or ordinal scale

do not make assumptions about the shape of population distributions

used to evaluate hypotheses about the shape of a distribution, rather than the distribution's mean, variance etc.

Less Powerful; more likely to reject a false null hypothesis

Front 17

Tests for Nominal Data

Back 17

1)Chi Square test--used to analyze the frequency of observations in each category (level) of a nominal variable

Single sample chi square test(also known as the goodness of fit test)

Multiple sample chi square test

Front 18

Degrees of freedom for chi-square tests

Back 18

tests for nominal data

single-sample chi square = categories - 1

multiple sample chi square =
(c-1)(r-1)

c=number of columns
r=number of rows

Front 19

Tests for Ordinal Data

Back 19

1)Mann-Whitney U Test
2)Wilcoxon Matched-Pairs Signed-Ranks Test
3)Kruskal-Wallis Test

Front 20

Mann Whitney U Test

Back 20

nonparametric test for ordinal data

Use: One IV with two independent groups; One DV that is rank ordered

The nonparametric ALTERNATIVE for the T TEST FOR INDEPENDENT SAMPLES

Front 21

Wilcoxon Matched-Pairs Signed-Ranks test

Back 21

nonparametric test for ordinal data

USE: One IV with two correlated (matched)groups; one DV with rank ordered data

the nonparamentric alternative to T-TEST FOR CORRELATED SAMPLES

the statistic=T

Front 22

Kruskal Wallis Test

Back 22

nonparametric test for ordinal data

Use: One IV with two or more independent groups; one DV with rank ordered data

the NONPARAMETRIC ALTERNATIVE to a one-way ANOVA

The statistic=H

Front 23

Tests for Interval and Ratio Data

Back 23

1) T-test (student's t-test)
2)Analysis of Variance (ANOVA)

Front 24

T-Test versus ANOVA

Back 24

A t-test is used to evaluate hypotheses about the DIFFERENCES BETWEEN TWO MEANS

whereas

an ANOVA is used to COMPARE TWO OR MORE MEANS (it helps control experimentwise error rate, decreasing prob of making a Type I error)

Front 25

T test for a single sample

Back 25

used when the study includes only one group and the group (sample) mean will be compared to a known population mean (In essence, the population is acting as a no-tx control group)

Use: One IV, single group
One DV with interval or ratio data

df=number of subjects - 1

n-1

Front 26

T test for independent samples

Back 26

used when a study includes two independent (unrelated) groups and the means of the groups will be compared

Use: one IV with two independent groups, one DV with interval or ratio data

df=n-2 or the total number of subjects minus 2

(or think, n of group 1 minus 1 plus n of group 2 minus one)

Front 27

T test for correlated samples

Back 27

used when the two means to be compared come from correlated groups (e.g., within subjects design or groups that used matching)

use: one IV with two correlated groups; one DV with interval or ratio data

df=number of pairs of scores minus 1

(e.g., may have 50 subjects, with 25 in each group that were matched...25-1 OR
25 people who are assessed before and after the IV...25 pairs of scores -1)

Front 28

One-way ANOVA

Back 28

USE: One IV with two or more independent groups (usually three or more b/c with two usually use t-test); one DV with interval or ratio data

df= (c-1), (n-c)

c=number of levels of the IV
n=number of subjects

Front 29

Factorial ANOVA (two-way, three way etc)

Back 29

USE: TWO OR MORE IV with independent groups; one DV with interval or ratio data

Front 30

MANOVA

Back 30

Use: one or more IV, and TWO OR MORE DVs

increases statistical power by simultaneously assessing the effects of the IV on all of the DVs

Front 31

ANCOVA

Back 31

combines and ANOVA with regression analysis, allows the investigator to control an extraneous variable by statistically removing the portion of variability in the DV that is due to the extraneous variable

reduces with-in group variability , resulting in MORE POWER

Front 32

Pearson product moment (r)

Back 32

type of correlation coefficient

used with both variables are interval or ratio data

Front 33

Spearman (rho)

Back 33

type of correlation coeff. aka spearman rank-order

both variables are rank ordered data

Front 34

Point biserial

Back 34

type of correlation coefficient

variable 1 is nominal data that rep a TRUE dichotomy (eg., male or female)

variable 2 is interval or ratio data

Front 35

Biserial

Back 35

correlation coeffecient

Variable 1 is nominal data reflecting an artificial dichotomy (e.g, favorable or unfavorable)

Variable 2 is interval or ratio data

Front 36

Eta
(!!!!!)

Back 36

a correlation coefficient used to ASSESS NONLINEAR RELATIONSHIPS

both variables must represent interval or ratio data

example studying the effects of anxiety on performance (too low or too high might equal poor performance, in the middle produces the best results)

Front 37

Three assumptions of Peason Product Moment (and most other correlation coefficients)

Back 37

1) linearity (linear relationship between variables)
2)unrestricted range (data collected from people who are heterogenous with regard to the characteristics)
3)homoscedasticity (range of Y scores is about the same for all values of X)

Front 38

Coefficient of Determination

Back 38

a squared correlation coefficient , that is interpreted as the proportion of variability in Y that is associated by X

Front 39

Multivariate Techniques

Back 39

used to assess the degree of association among three or more variables and to make predictions that involve, at a minimum, two predictors and one crieterion

Front 40

Multiple Regression

Back 40

a type of multivariate technique

used when two or more continuous or discret predictors will be used to predict status on a single continuous criterion

The output is a MULTIPLE CORRELATION COEFFICENT (R)

May be used in place of ANOVA when groups are unequal in size

Front 41

Simple (simultaneous) Regression

Back 41

analyzing the effects of all of the predictors on the criterion at once

Front 42

Forward or Step up Regression

Back 42

One predictor is added in each subsequent analysis

Front 43

Backward or Step Down

Back 43

all predictors are used, then one predictor is eliminated in each subsequent analysis

Front 44

Multiple Regression vs. ANOVA

Back 44

-Mult. Regression is better when groups are unequal in size, b/c this can reduce the power and robustness of the ANOVA

-use MR when the IVs are measured on a continuous scale, b/c an ANOVA would require that the continuous data be broken into categories or levels which reduces power
-MR permits a research to add or subtract IVs (predictors) to the analysis to determine which subset best explains the variability in the DV (criterion)

Front 45

Canonical Correlation

Back 45

Type of Multiple Regression used when two or more continuous predictors (IVs) are used to predict the status on two or more continuous criteria (DVs)

Front 46

Discriminant Function analysis

Back 46

Type of multiple regression used when two or more continuous predictors (IVs) are used to predict a person's status on a single discrete (nominal) criteria (DV).

Front 47

Causal Modeling

Back 47

1)Path Analysis (Structural Equation)
2)LISREL

test a predefined causal model or theory

CANNOT PROVE causality, but can provide evidence that the causal theory or model is correct or incorrect