Exam 2 Flashcards
Scales of Measurement
Nominal, Ordinal, Interval, and Ratio Scales
- Importance: the different scales allow data to be analyzed and processed in different ways
Nominal Scale
Classification/categorization of objects/individuals
- Order doesn’t matter
- No numerical meaning
Examples
- male vs. female
- ethnicity
- favorite color
Ordinal Scale
Classification/categorization of objects/individuals where the order of categories has meaning
- Unequal differences between categories, intervals aren’t necessarily the same
Examples
- Placement in a race
- Ranking favorite ice cream flavors
- Grades
Interval Scale
Categorization of something where order matters and scales of measurement are equal
- Don’t have true zero value
- True zero value: when zero means the nonexistence of the measure (such as 0 people in a population would be an absence of people)
Examples
-Temperature
Ratio Scale
Categorization of something where order matters and scales of measurement are equal, and where there is a true zero
- The ratios are meaningful (eg. x is 3x as much as y)
Examples
- Height
- Age
- Weight
Measures of Central Tendency (Averages)
Mode (bimodal and multimodal), median, mean, and outliers
Mode
The score that occurs most often
- Can be used with any scale
- Only average that can be used for nominal scale data
Bimodal: two scores occur equally often + most frequently
Multimodal: three + scores occur equally often + most frequently
Median
The middle point in a set of scores
- 50% of scores fall above this point, 50% below this point
- Can be used for ordinal, interval, and ratio scales
Mean
The arithmetic average found by adding all scores and dividing by the number of scores
- Can be used for interval and ratio scales
- May be biased by outliers
Outlier
Extreme values much higher/lower than the majority of scores
- Greatest effect on mean
- Wont largely impact median and mode
Measures of Dispersion
How spread out data are
- Includes range (inclusive and exclusive), standard deviation, and variance
Range
Distance from highest to lowest value
Inclusive: Highest score - lowest score + 1
Exclusive: Highest score - lowest score - 1
- Outliers ruin
Standard Deviation
Expresses average distance of scores from the mean
- Only calculated on interval/ratio data
Variance
Standard deviation squared
- Only calculated on interval/ratio data (requires mean)
Correlation
The degree of a relationship between 2 variables
- Range between -1.00 - 1.00
- Correlation does NOT mean causation
Strength
- absolute value of correlation
- - .90 > .70, .90 > .70
Direction
- positive: when one variable increases, other also increases (and vice versa)
- negative: when one variable decreases, the other increases (and vice versa)
Bivariate Correlation Coefficient
Correlation coefficient is referred to as this when the relationship between two variables is being assessed
Types of Correlation
Pearson’s product-moment correlation (Pearson’s r)
- used with interval/ratio data
Spearman’s rho (p)
- if one or both variables are ordinal
Multiple Correlation
Correlation between multiple variables and one particular variable
- Single score
- 0.00 - 1.00
Multiple regression
- similar but gives info about each predictor variable (individual contributions)
Limits of Correlations
-There a different kinds of correlations for different kinds of data
- It is only evidence of a relationship not a cause
Error Variance
Natural fluctuations in group caused by something other than independent variable
- Differences within a group
- one measure = standard deviation
- Can be used to find significant difference by completing the equation: differences between groups/differences within groups = effect of independent variable + error variance/error variance
T-Test
Used to compare two groups to determine if there is a significant difference between them
Analysis of Variance (ANOVA)
Used to compare three+ groups to determine if there is any significant difference
Between-Groups Variance
Measurements found by testing different groups on different variables
Pros: Prevents carryover effect
- Shorter duration
- Reduces impact of variables
- Results are easier to interpret
Cons: Requires more participants
- Individual differences can cause results
Within-Groups Variance
Measurements found by testing the same group/individual on different variables
Pros: increased statistical power
- Control for individual differences
- Potential reduced error-variance
Cons: Possibility for order effects
- Carryover effects
- Practice effects
- Time measurement effects
Parameter
Characteristic of a population
Parametric tests
Makes assumptions about population characteristics
- Only usable with interval/ratio data
Nonparametric tests
Makes no assumptions about population characteristics
Experiment
Researcher manipulates independent variable to see if there are any differences in the dependent variable
- Among equivalent groups
- Use random assignment
- Yields causal information about the effects of an independent variable (when done correctly)
Quasi-Experiment
Similar to an experiment as an independent variable is manipulated BUT groups aren’t equivalent
- Could be due to lack of random assignment, etc.
- May yield causal information if confounds are eliminated
Correlational Study
Explore existing effect of subject variable on dependent variable
- Cannot yield causal information
- Identify relationships between subject variable and dependent variable
- Sometimes only option
Experimental Group
Group experiencing manipulation
- For example, administered real drug
Control Group
Group that doesn’t experience manipulation OR experiences placebo
- For example, given sugar pill
Placebo
Inert treatment with no effect on dependent variable
- Helps counteract demand characteristics
- Sugar pills or distractor task
How can you ensure that independent variable manipulation is causal?
- Groups must be equivalent before introducing independent variable
- Must introduce IV before measuring dependent variable
- Must be free of confounds
Random Assignment
All participants have an equal chance of being assigned to any group in an experiment
- Can be achieved through: coin flip, pulling names out of a hat, random # table
- Not the same as random selection
- Not a guarantee but should lead to equivalency with enough participants
Selection Bias
Occurs if researchers choose the participants groups, or if participants choose their own group.
- Can lead to inequivalent groups
- Any difference between groups should be due to random choice in random assignment
Matching
Identifies pairs (or triplets, quadruplets, etc.) of participants who measure similarly on a characteristic related to dependent variables and then randomly assigns each of the participants to a different experimental condition
- Equivalent on 1+ important characteristic
- Good with limited subject pool, not good enough for good random assignment
Downsides
- Can be difficult to find good matches
- Only controls for matched characteristic
Pretesting
Essentially the same as the actual experiment
- Establish a measure related to the dependent variable
- Not always practical as it can cause ceiling effect (could raise all scores too much)
Subject Variable
Characteristic of participants that can’t be changed
- Makes random assignment impossible
Examples
- Gender
- Age
Cross-Sectional Design
Examines differences across age groups
- Different individuals for different age groups
Extraneous Variables
Variables other than the independent variable that can effect the dependent variable
- Must be equivalent across groups
- Causes confounded results when the extraneous variables change with the independent variable
- Can be controlled through balance and holding constant
- AKA confounds
Confounds
Extraneous variables or any other flaw in research
- Fewer confounds = more internal validity
Examples
- Experimenter Bias
- Demand Characteristics
- Instrumentation Effects
- Subject Attrition
Experimenter Bias
Any confound caused by research expectations
- Interpretation of ambiguous data
- Treating experimental/control groups differently
Demand Characteristics
Any confound caused by participant expectations
- “What does experimenter hope to find”
Instrumentation Effects
Occurs when instrument used to measure dependent variable changes in accuracy over time
- Machines/tools wear out
- People learn
- People get tired
- Leads to differences in how each participant is measured
Subject Attrition
Participants may leave study partway through
- Nonsytematic: leaving for reasons unrelated to independent variable, may not threaten internal validity if it is more or less equal across conditions
- Systematic: participants leave study in unequal numbers from different groups, possibly related to independent variable, severe threat to internal validity
Single Blind Procedure
Participant doesn’t know condition
Double Blind Procedure
Experimenter + participant don’t know condition
How can you ensure internal validity?
Ensure different groups experience almost the same circumstances
- Time
- Location
- Researchers
- What they’re told
- Weather
- How researcher is dressed
- Etc.
Types of Within-Subject Designs
Pretest-posttest design, repeated measures design, longitudinal design
Pretest-posttest Design
Two measures of dependent variable for each participant
- One before independent variable manipulation
- One after
Repeated-measures Design
Multiple dependent measurements for each participant
Longitudinal Design
Repeated-measures design that occurs over an extended time (days, weeks, months, years)
- Often paired with cross-sectional studies
Benefits of Within-Subjects Design
Require fewer participants
- Each participant provides a score or scores for each level of the independent variable
Lower error variance
- No difference between groups
- Same variability “should” hold for different levels of independent variable
Disadvantages of Within-Subjects Design
Particularly susceptible to demand characteristics
- More opportunities to guess study purpose
- Expectation to do “differently” in different conditions
- May require use of “ethical” deception (eg. placebo)
Carryover Effect
- Practice Effects
- Fatigue Effect
- Counterbalance
- History Effect
- Maturation Effect
- Testing Effect
- Regression toward the mean
Cannot test subject/selected variables
- Despite flaws, proper control can compensate for many disadvantages just not the last one
Practice Effect
Get better at task each time
Fatigue Effect
Get worse at task each time
History Effect
Something happens during study that effects dependent variable measures
Maturation Effect
People and their scores may naturally change over time
Testing Effect
People improve on tests with multiple takings
- Independent of anything else
Regression toward the mean
Tendency for extreme scores to normalize when retested
- Independent of manipulation
- Common when participants are selected based on pretest
Counterbalancing
Present different experimental conditions to participants in different orders
- Controls for many carryover effects
- May be complete/incomplete counterbalancing
Complete Counterbalancing
All subjects experience each condition several times until they have experienced every possible order
- Any carryover effects should influence all conditions equally
- ABBA Counterbalancing
- Block Randomization
ABBA Counterbalancing
- Used w/ two experimental conditions
- Ensures first condition is also the last
- May be repeated as often as necessary
- Potential Issues: linear improvement vs. nonlinear practice effects, may lead to issues if participants notice patterns
Block Randomization
Each block consists of a single presentation of each experimental condition in a unique order
- Presented in random order
- Useful for 3+ conditions
- Need enough blocks for randomization to work
- Each condition should be first/last a roughly equal number of times
Incomplete Counterbalancing
Participant receives unique order of all conditions at least once
- Does not receive all possible orders
- Complete design isn’t always practical, leading to incomplete counterbalancing
- Must have enough participants to counterbalance potential fatigue + practice effects
- Random Order with Rotation
- Balanced Latin Square (Latin Square)
Random Order with Rotation
Establishes random order and use that order with first participant, move first condition to last condition for each participant
- eg. DBCAE to BCAED to CAEDB
- Make sure # of participants is a multiple of possible orderings
Balanced Latin Square (Latin Square)
Each condition is presented in each possible position, and presented before/after each other condition
- Even # of conditions: same # of orderings as conditions
- Odd # of conditions: twice as many blocks (5 conditions = 10 orderings)
- Choose # of participants that is a multiple of the # of orderings
What are the limitations of counterbalancing?
- Amount time
- # of participants
- Permanent changes