Research & Stats Flashcards
Moderator Variable
Affects direction and strength of the relationship between IV and DV.
(mOderator is Outside the relationship of IV and DV)
Mediator Variable
Explains the relationship between IV and DV
(mEdiator Explains the relationship between IV and DV)
Bar graphs
-nominal or ordinal data
-categories on X axis
-percentage / recorded data on the Y axis
Histogram
-Interval or ratio data
-score listed in order on X axis
-number/percentile of the observation on Y axis
Frequency Polygon
-line graph
-interval or ratio data
Normal Distribution
-symmetrical
-central measures of tendency are equal to same value
-68% scores fall within 1 SD
-95% scores fall within 2 SD
-99% scores fall within 3 SD
Positively Skewed
-Tail at the higher end of the X axis
-Mean is largest central tendency
-Mode is smallest central tendency
Negatively Skewed
-Tail at the lower end of the X axis
-Mean is smallest central tendency
-Mode is the largest central tendency
History
Threat to internal validity
-events that occur during the study (effect results)
*Control
-more than 1 group
-random assignment
Maturation
Threat to internal validity
-physical, cognitive and emotional changes that occur in the subjects (due to passage of time).
*Control
-more than 1 group
-random assignment
Differential Selection
Threat to internal validity
-different assignment to treatment groups
-when groups differ at the beginning of the study due to assignment.
*Control
-random assignment
Regression towards the Mean (aka Statistical Regression)
Threat to internal validity
-participants selected for their extreme score on pre-test gradually start to shift towards mean scores later on in the study (post test assessment).
-characteristics assessed for are typically not stable over time.
Testing
Threat to internal validity
-taking a pretest effects later responses
*Control
-no pretest
-Soloman four group design
Instrumentation
Threat to internal validity
-when instrument used to measure IV changes over time.
Differential Attrition
Threat to internal validity
-participants dropout of 1 group
-composition of groups are altered
Reactivity
Threat to external validity
-when participants respond differently to the IV during a study than they would normally.
*Control
-single or double blind study
-deception
-unobtrusive measures
Multiple Treatment Interference
Threat to external validity
-carryover effects and order effects
*Control
-counter balancing (different groups receive different levels of IV)
-Latin Square Design (type of counter balancing)
Selection Treatment Interference
Threat to external validity
-research participants differ from the population of interest.
*Control
-random selection
Pretest Treatment Interactions
Threat to external validity
-when taking a pretest effects how participants respond to the IV.
Soloman Four Group Design
-used to identify the effects of pretesting on a study’s internal and external validity
-4 groups
-2 levels (2 groups exposed to IV, 1 of the 2 groups are given a pretest)
Population validity
External validity
-how generalized are finding to the population participants are pooled from.
Ecological validity
External validity
-how generalized are findings in different settings
Temporal validity
External validity
-how generalized are findings across time
Treatment validity
External validity
-how generalized are finding given variation of the treatment (IV)
Outcome validity
External validity
-how generalized are findings to different but related DV
Ground theory
-develop a generalized abstract theory of process, action or interaction, coming from the views of participants.
*interview and observation
Phenomenology
-gain in depth understanding of lived experience of participants
-how participants perceive, describe, feel, judge, remember or discuss a observed subject.
*interview
Ethnography
-studying participants in natural culture or setting
*participant observation (joining a culture and paticipating)
Thematic Analysis
-identifying, analysis and reporting patterns within the data.
-is stand alone or starting point for other methods.
*interviewing and focus groups
Probability sampling
-random selection of the sample from the population
*prone to sampling error
Systemic random sampling
-when random list of all individuals in the population is available.
-selecting every 10th, 20th, 30th name.
Stratified random sampling
-when population is heterogenius with regard to 1 or more characteristics (gender, age, diagnosis)
-researcher wants to make sure every group is represented within the category. (9th graders, 10th graders, 11th graders, 12th graders - high school students)
Cluster random sampling
-randomly selecting clusters and then including all of those in the clusters or random selection from each cluster.
Non-probability sampling
-all members of the population do not have equal chance of being selected.
*vulnerable to sampling error and bias
Range of correlation coefficient
-1.0 to 1.0
correlation coefficient assumptions
- relationships between variables is linear
- unrestricted range of scores for all variables
- there is homoscedasticity
Persons r
-both continuous variables
-interval or ratio
Spearman rho
-when the variables are reported as ranks (ordinal)
Point Biserial Correlation
-when one variable is continuous and the other is a true dichotomy (nominal variable with only 2 categories)
Biserial Correlation
-one variable is continuous the other is an artificial dichotomy (when the continuous variable is dichotomized)
*e.g., final exam scores when a cutoff score is used to create 2 categories (pass and fail).
Contingency Correlation
when both variables measure on a nominal scale
Coefficient of Determination
*measure of shared variability
*amount of variability in one variable that explained by the variability in the other variable.
e.g., Prediction (job knowledge)
Criterion (job performance)
*Correlation coefficient is .70
-the square root of .70 is .49
-this means that 49% of variability in job performance is explained by differences in level of job knowledge while 51% is due to other factors.
Multiple Regression
-2 or more predictors
- 1 criterion.
-continuous scale
- Simultaneous
- Stepwise
*desired outcome is high predictors have high correlation with criterion and low correlation with other predictors.
Multicollonearity
when predictors are highly correlated with other predictors.
Canonical Correlation
- 2 or more continuous predictors
-2 or more continuous criterion
Discriminant Function Analysis
-2 or more predictors
-1 criterion
-nominal scale
*alternative is a Logistic Regression
Central Limit Theorem
*estimate the characteristics of the sampling distribution
- sampling distribution will get closer to a normal shape as the sample size increases.
- the mean of the sampling distribution means will be equal to the population mean.
- the SD of the sampling distribution (standard error of means) will be equal to the population SD divided by the square root of the sampling size.
Retain a true null
-correct conclusion
-IV has no effect
Reject a false null
-correct conclusion
-IV has an effect
Alpha
-probability of making a type I error
level of significants .05 or .01
Beta
-probability of making a type II error
Reject a true null
Type I error
Retain a false null
Type II error
Increasing alpha
.01 to.05 or .05 to 1.0
-increases likelihood of a Type I error
-decreases likelihood of a Type II error
Decrease alpha
.05 to .01
-Decrease likelihood of a Type I error
-Increase likelihood of a Type II error
Statistical power
ability to reject a true null hypothesis
Single Sample Chi-square
-descriptive study
-1 variable
-nominal data
Multiple Sample Chi-square
-descriptive study
-2 or more variables
Single-Sample t Test
-compare an obtained sample mean to a known population mean.
-1 IV (2 levels)
-1 DV
-Interval or ratio
Independent Sample t Test
-compare the means obtained by 2 groups
-subjects in the groups are unrelated
-1 IV (2 levels)
-1 DV
-Interval or ratio
Paired Sample / Correlated Samples t Test
-compare 2 means when there is a relationship between subjects in the 2 groups.
*natural pairs or within subjects (paired with themselves)
-1 IV (2 levels)
-1 DV
-Interval or ratio
One-Way ANOVA
-1 IV (more than 2 levels)
-1 DV
-interval or ratio
F ratio
Mean square between (MSB) - variability in DV scores due to treatment and error
Mean square within (MSW) -variability that is due to error only
MSB/MSW = F ratio
*When F is larger than 1.0 the IV has an effect
Factorial ANOVA
- More then 1 IV
-1 DV
-produce separate f ratio for the main effect of each IV and their interaction effect
Mixed ANOVA
-at least 1 IV between subjects
-at least 1 IV within subjects
-1DV
Randomized Block ANOVA
-used to control for effects of extraneous variables
-used the extraneous variable as a IV
-determine main effect and interaction effect
ANCOVA
-control effect of extraneous variable on a DV by using regression analysis
-statistically removing its effects from the DV
MANOVA
-1 or more IV
-2 or more DV
Cohen’s d
-measure effect size
-indicates differences between two groups in terms of SD.
Calculated: dividing mean difference by the groups on the IV by the pooled SD of the 2 groups.
d less than .2 = small effect size
d between .2 and .8 = moderate effect size
d larger than .8 = large effect size
Alternative to Pearson’s r
Eta
When variable relationship is non-linear when using Pearson’s r
Pearson’s r will underestimate the relationship between variables.
How do you calculate the Coefficient of Determination?
Square the correlation coefficient
Alternative to Discriminative Function Analysis
Logistic Regression
Mean Square Between
Variability in the DV due to treatment and error
Mean Square Within
Variability in the DV due to error alone
How to calculate F ratio
MSB/MSW
Significant F ratio is
> 1.0
Trend Analysis
1 or more IV
*Determine if there is a linear or nonlinear relationship between IV and DV.
Planned Comparison (Priori Test)
Type of post-hoc used when:
there are 2 t-test or more
Post Hoc Test
When an ANOVA produces a significant F ratio
*use of t-test to compare groups
Bronferroni Procedure
Used to control experiment wise error rate
*dividing alpha by the number of significant statistical test to obtain an alpha level for each test.
Alternative to Cohen’s d
Cohen’s f - when there is more than 2 groups
Jacobson Traux Method
evaluating clinical significants for each participant in a study
“orthogonal” means:
uncorrelated
The size of the standard error of the mean increases as:
the population standard deviation increases and the sample size decreases.
