Statistics and Research Flashcards
Research
Classified as:
- Qualitative (relies on unstructured or non-numerical data)
- Quantitative (experimental or non-experimental - relies on numerical data)
Analogue Study
Conditions are in some way an analogue (approximation) of actual clinical practice.
Variables
Behaviors/Characteristics that vary from one person to another and from one situation to another
Independent Variable (Cause)
What you change
Dependent Variable (Effect)
What changes as a result of that ^
Mediator Variables
Explains or accounts for (is responsible for/causes) the relationship between the IV & DV
Moderator Variable
Affects the direction or strength of the relationship btwn IV & DV
Behavioral Sampling
Recording a specific aspect of behavior
Behavioral Sampling Techniques
- Interval Recording
- Time Sampling
- Latency Recording
- Duration Recording
Event Recording
Best for behaviors that:
- occur infrequently
- have long duration
- Leave a permanent record or other product
Latency Recording
Used to determine how long it takes for a behavior to begin after specific event
Interval Recording
Useful when target behavior has no clear beginning or end
Duration Recording
- When behavior has a clear beginning and end
- Indicates how long a behavior lasts
Situational Sampling
Observing behaviors in multiple settings
Sequential Analysis
- Used to encode behavior sequences
- Useful for studying complex social behaviors
Simple Random Sample
Every member of the population has an equal chance of being selected for inclusion
Cluster Sampling
Selecting units (groups) of individuals rather than individuals from the population (pre-existing groups from schools, mental health clinics, etc.)
Event Sampling
Behavioral sampling that involves observing and recording information about a behavior when it occurs
Controlling Extraneous Variables (EVs)
Matching - when # of subjects too small
Blocking (grouping) - EV treated as IV
ANCOVA - used to statistically remove effects of EV on the DV
Internal Validity
When adequate:
- Conclude that observed variations in DV are due to variations in IV & not other factors
Campbell & Stanley
(Generic EVs that limit study’s Internal Validity)
(Factors affect the DV)
- Maturation (changes w/in subj. due to passage of time)
- History (external event during course of study)
- Testing
- Instrumentation (change in tests or other measuring devices)
- Statistical Regression (v. high/v. low scores go towards mean on retest)
- Selection (subj in diff groups not similar)
- Attrition (Mortality) - subj. who drop out from 1 grp diff. from subj. in another grp)
- Interactions with selection (e.g. w/hx say one grp exposed to event & other is not)
External Validity
Generalization of results to other people, settings and conditions
Pretest Sensitization
- Administration of a pretest affects how subj. react to the treatment (Use Solomon-Four Grp Design to evaluate pretest sens.)
Threats to External Validity
- Interaction btwn selection & treatment
- When pple in sample differ from pple in population (respond differently to IV)
Threats to External Validity
Reactivity
- Occurs when research participants act differently because they know their behavior is being observed
Other Threats to External Validity
- Experimenter expectancy (experimenter may bias result of a research study)
- Demand Characteristics (cues in research setting that may communicate to subj what is expected of them)
- Multiple Tx Interference (more than one level of IV is administered to each subj) - :Corrected by Latin Sq Design
Latin Square Design (External Validity)
Partial Counterbalancing design
- determines what sequence of tx will be administered to diff. groups of participants
Counterbalancing Design
Goal is to use every possible sequence of tx
Within-Subjects design
- administering diff. levels of IV to diff. groups in a diff. order
Group Designs
- Between-Groups
- Within-Subjects
- Single-Group Time Series
Between Groups Research Design
different people test each condition, so that each person is only exposed to a single user interface
An experiment that has two or more groups of subjects, each being tested by a different testing factor simultaneously
Within-Subjects Research Design
the same person tests all the conditions (i.e., all the user interfaces
All participants are exposed to every treatment or condition. The term “treatment” is used to describe the different levels of the IV, the variable that’s controlled by the experimenter
Carryover Effects
Occur when being exposed to 1 level of the IV affects how a participant reacts to another level of that IV (Counterbalancing controls for this - present diff levels of IV to diff. participants in a diff. order).
Single-Group Time Series Design
DV is measured several times before & after IV is applied
Factorial Design
Research design with more than one IV
Main Effect
Effect of one IV on DV
Interaction
Effect of one IV on DV at different levels of another IV
Static Group Comparison
Two intact groups (tx & no tx
- evaluate effects of tx by comparing post-test scores of the two grps.
Single Subject Design (Baseline & Treatment phases)
- AB design (1 baseline & 1 Tx phases)
- Reversal Design (extends AB @ min. 2 baseline phases & 1 Tx)
- Multiple baseline design: apply IV to 2+ baselines e.g. 2 settings, tasks/behaviors) - ABAB
Scales of Measurement (NOIR)
- Nominal (name, color, gender)
- Ordinal (good/bad; tall/short; hi/l)
- Interval (temperature)
- Ratio (grades)
Peaked Distribution
Leptokurtic (think leap)
Flat Distribution
Platykurtic (think duck-billed platypus’ bill)
Negatively Skewed Distribution
(Tail on +ve side of central value)
Tail Tells the Tale
- When scores are concentrated on the positive side (tail) of the distribution
- The mean of negatively skewed data will be less than the median, which is less than mode
Positively Skewed Distribution
Tail on -ve side of central value
- When scores are concentrated on the negative side (tail) of the distribution
- The mean of positively skewed data will be greater than the median, which is greater than mode
Measures of Variability
- Range (subtract lowest from highest)
- Variance (spread of #’s in a data set)
- Std Deviation (sq rt of variance)
Constant Added or Subtracted
constant does not change
- Measures of Central Tendencies (mean, median & mode) change
- Measures of Variability (range, variance & std deviation) stay the same
Constant Divided or Multiplied
All measures of central tendencies & measures of variability change
Rejection Region (region of unlikely values)
Alpha (0.1 or 0.5)
0.1 = 1% falls in region of unlikely values & 99% fall in region of likely values
Type I Error = Alpha (level of significance)
True Null Hy is rejected
- accept that IV has effect on DV when it is due to sampling error
Type II Error = Beta
False Null Hy is retained
- accept IV has no effect on DV when it actually did
Caused by IV not administered in sufficient intensity/length of time
- sample size too small
- alpha too small
Power
- alpha is increased
- sample size increased
- maximized when use one-tailed/ t-test/ANOVA/parametric test
Parametric Test (inferential stat tests)
Used:
- When data being analyzed rep. interval or ratio scale of measurement
- Population is normally distributed
- Homoscedasticity is met: equal variance from diff. grps rep.
Critical Value (cut-off point) (inferential stat tests)
Determined by:
- alpha
- degrees of freedom
Test for Nominal Data
- Chi-square: very sensitive to sample size
- Single-sample chi-square (1 IV & data is in frequencies)
- Multiple sample chi-square (2+IVs & frequencies in each nominal category)
Tests for Ordinal Data
- Mann-Whitney U:1 IV, two unrelated grps
- Wilcoxon Matched-Pairs: 1 IV grps related
- Kruskal-Wallis: rank-ordered data 2+ grps
Tests for Interval & Ratio Data
- T-test single sample (compare known obtained grp/sample mean to known population mean)
- T-test ind. samples (compare means in 2 grps that are indepe. (diff)/unrelated
- T-test Correlated samples (compare means in 2 grps that are related)
- One-Way ANOVA (1 IV & 2+ indepe. grps)
- Factorial ANOVA (2+IVs)
F-Ratio
MSB(reflect tx effects & error) /MSW (error)
MSB larger than MSW f-ratio is > 1
Measures of Effect Size
- Cohen’s d
- Eta squared
Eta
- Used to determine degree of association when relationship is nonlinear and two continuous variables)
Correlation Coefficient (Avg. degree of association btwn variables)
Pearson r - both variables interval or ratio
Spearman rho - both variables are ranks
Tetrachoic - 2 normally distr. continuous variables
Point biserial - 1 true dichot/1 ratio/interval
Biserial - 1 artificial dichot/1 interval/ratio
Bonferroni Test
- Reduces level of alpha for each comparison thus reducing probability of Type I Error (replaces ANOVA & Post-hoc)
Multiple Regression
- used when predictors & criterion are continuous
- “Dummy coding” use when you have categorical data
