Research Design Flashcards
Types of Research Designs
Experimental
- true experiment
- quasi-experiment
or Non-Experimental
- correlational
- descriptive/observational
True Experiments
→ highest level of internal validity, can make the strongest causal claims with a true experiment, but we lose some external validity
Requirements for a True Experiment
- Systematic manipulation of 1 or more variable(s) between or within groups
- Guarantee temporal order of cause effect
- Observe covariation between variables
- Minimise alternative explanations/confounds - through random allocation
- Random assignment to each condition/group
- Minimise alternative explanations/confounds
Advantages of a True Experiment
- Isolation of a single variable
- Systematic manipulation of the variable (operationalising, choosing controls, etc means more control over determining if the variable is really the thing that might cause it)
- Doing so allows us to minimise numerous alternative explanations
- High levels of internal validity
Random Assignment
randomly assign participants to each of the groups to reduce the likelihood of systematic differences between the participants in the group which undermine internal validity
- Can be sure we have eliminated confounds
- controlling for bias in group allocation
- internal validity
Random Sampling
approach to recruiting subjects for your study
- Try to sample different elements of the population proportionally
- More representative
- Applies to all forms of research design
- External validity
- c
Two-way design
most common, two IV’s for example video games and also the duration of play
- Helps look at the difference between a wider range of combinations to determine what contributes to this
- Can examine the difference between the groups and any interaction
- Is there a difference between the amount of time played
- Is there a difference between the type of game
Advantages and Limitations of within-subject design
Advantages
- Can be statistically powerful - remove error noise
- Easier to get a significant result
- Accounts for individual differencesLimitations
- Fatigue
- Practice
- Carry-over
LSD and Psychopath Psychotherapy
- Treatment of psychopaths in prison
- One reason it’s hard to treat psychopaths for violent crimes is that they lack empathy, LSD in high doses creates an “ego death”
- “First-ever marathon nude psychotherapy session for psychopaths. Raw, naked, LSD sessions for eleven days straight
- Theory that it would foster a sense of empathy
- Obviously didn’t work
Follow-up in the 80s
- Normal (recidivism) rate of offending is 60%
- Those with therapy is 80%
Characteristics of Quasi Experiment
- Research only has partial control over the independent variables
- Participants are assigned to groups or conditions without random assignment
Two types
- Person x treatment
- Natural experiments → less relevant for psych
- Useful when random assignment is not possible or ethical or when researchers don’t have control over what is being measured (race, ethnicity, age)
- use dependent and true independent variables but also use quasi-independent variables
Quasi Independent Variables
- not manipulated by the experimenter
- random assignment is not possible
- used in the same way as a true independent variable
- split people into groups based on the variable
Person/Attribute Variables
Individual difference variables
- Can vary along a spectrum
- Can be based on diagnostic criteria
- Use these variables most commonly for comparing groups – grouping variables
- We can use these to compare any differences on a dependent variable when random assignment is not possible
- Used because we can’t control in any other way
- Must be measured prior to the experiment, if not there are issues with internal validity, require temporal sequence to make sure the experimenter didn’t cause the difference
Extroversion vs Introversion Scale (Attribute Variable Example)
- Select random group of participants, complete a personality test which measures a range of personality traits
- Examine the scores and then select a group that score high in extroversion and introversion
- The attribute is measured and then participants are split into groups on the basis of their score
How do we split
- Splitting these attribute variables into high and low is a common practice, past q1 and q3 for example
- Not the best method statistically (quite crude)
Attribute of interest: usually normally distributed in the population
Natural Variables
- country of birth
- biological sex
- age
These cannot be changed but instead you can group someone by these categories
Non-Experimental Research Designs
No manipulation only measurement
- correlation
- descriptive/observational
Experimental Research Designs
Manipulated and measurement variables
- true experiment
- quasi-experiment
True Experiments (and their positive aspects)
It has the highest level of internal validity, it can make the strongest causal claims with a true experiment, but it lacks external validity
- It is very hard in the real-world to actually isolate the variable that may cause something, and so true experiments are good because through deduction you can start to pick out these variables and determine their effects
Requirements for a True Experiment
- Systematic manipulation of 1 or more variable (s) between or within groups
- To guarantee temporal order of cause effect
- To observe covariation between variables
- To minimise alternative explanations/confounds - through random allocation
- Random assignment to each condition/group
- To minimise alternative explanations/confounds
Advantages of a True Experiment
- Isolation of a single variable
Systematic manipulation of the variable (operationalising, choosing controls, etc means more control over determining if the variable is really the thing that might cause it)
- Doing so allows us to minimise numerous alternative explanations
- High levels of internal validity
Justify a True Experiment for Testing whether Video Games cause Violence
when making a causal claim, a true experiment would best isolate the variables
- Independent variable = violent video games (must be operationalised, such as call of duty)
Dependent variable - aggression or violence (must be operationalised, going to lose external validity because it is immoral to measure this in the same way)
Correlating Violence to a Hot Sauce Task
- IV - COD
- DV - hot sauce allocation
- Control group - play a non-realistic violent game (angry birds)
- Experimental group - play COD
Random Assignment
In a true experiment it is important to randomly assign participants to each of the groups to reduce the likelihood of systematic differences between the participants in the group which undermine internal validity
- this can make it so that we are sure that we have eliminated confounds
Approach to Random Sampling Subjects for your Study
- Try to sample different elements of the population proportionally
- This is more representative
- Applies to all forms of research design
- Effective for improving external validity
One-way Design
One independent variable
- less common
Two-way Design
Two independent variables
- most common
- helps look at the difference between a wider range of combinations to determine what contributes to this
- Can examine the difference between the groups and any interaction
E.G
- Is there a difference between the amount of time played
- Is there a difference between the type of game
Within-subject design
Testing the same participants under multiple conditions or at different points in time
- Use the same participants in different conditions rather than different groups
- Allows someone to eliminate individual differences
- Sometimes called a repeated measures design
- There is a systematic manipulation of IV still
Advantages of a Within-subject Design
- Can be statistically powerful - remove error noise
- Easier to get a significant result
- Accounts for individual differences
Limitations of a Within-subject Design
- Fatigue
- Practice
- Carry-over
Counterbalancing
A potential solution to the limitations of a within-subject design
- repeated measures where the same group experiences all the levels of the IV
- it can be used with random assignment to potentially reduce order effects
Between-subject Design
Where participants are split into two or more groups and each group is assigned a treatment condition
- these treatments are then compared to the other groups to see differences in the outcome of the experiment
Ways of Splitting Attribute Variables
Median Split
- Find midpoint
Advantage
- Easy
- Keep all participants
Disadvantage
- Participant 10 and 11 are similar
- Loss of information about unique individual differences
Person/Attribute x Treatment Design
- Quasi-independent variable – measured not manipulated and no random assignment
- True-independent variable – manipulated and random assignment
- Dependent variable – measured by the experimenter
- Allow us to examine group differences and how they interact with a manipulated or treatment variable
Natural Variables as a Form of Quasi-Independent Variable
variables manipulated by nature
- don’t want to find these variables ad-hoc, rather it should be found beforehand
- can’t be manipulated by the experimenter or randomly assigned
- allows us to look at the effects of specific environments and experiences on the individual, such as in a war zone
Example of a Natural Variable in a Quasi-Experiment
- Are women better than men at science
Quasi IV - sex=natural variable
- Is this biological sex
- Determined by chromosomes alone?
- Must be operationalised
DV: performance on science tasks
- it is very difficult to isolate these issues which is the problem with third variables and quasi experiments - in relation to war there are many different circumstances that might change their relationship to the war
Natural Variables VS Attribute Variables
- natural variables can be measured in units, they are quantitative
- attribute variables are qualitative
Neuroscience Research with a Different Psychopathy Example -
- James Fallon running study with less controls than psychopaths
- So he got his family as controls
Scanned brains so many times that he ended up putting his brain in the psychopaths side
- The researcher ended up having the same neural imaging as these
- Psychopathy might also be a spectrum variable
- E.G a lot of successful people may be the same
Quasi-Experiment Example of Psychopathy
Psychopaths can be considered as both a natural and attribute variable
- they are first primed with ‘ready’ on the screen, and then are asked to respond to where the symbol appears
- a valid response is to press where the symbol is, and an invalid one is to press where the prime was
- the participants in the experiment were separated into groups and matched by their important variables, such as age or level of education
- IV: trial type
- DV: reaction time/error
- Found that higher errors in psychopath groups than prisoners in incongruent trials
- Psychopaths are argued to have issues with processing emotional information
- Suggestions of amygdala dysfunction in psychopaths
Patching
- Imcluding multiple control groups to cover all bases and adjust to any threats towards internal validity
- Patching tries to meet specific traits, matching tries to hold things as constant as you can
Threats to Quasi Experiments
In quasi experiments the lack of random assignment or controlled manipulation of the quasi independent variable means;
- we can never be certain of temporal order of quasi IV and DV
- third variables may be alternative explanations (pre-existing group differences on other variables, we can try match groups on other characteristics)
Basics of Correlational Research
relationship between variables (no causation)
- low internal validity
- sometimes have higher external validity although it is dependent on design (observational have the highest external validity)
How does Correlational Research Work?
Take a series of measurements
- Always multiple dependent variables
- Experimenter doesn’t manipulate anything, just measures the way participants change (do they change the same way, differently etc)
Correlational Research having similar issues Regarding Descriptive Studies
measurement/testing effects, question wording, random sampling need to ensure external validity (rather than random allocation in true experiments)
Characteristics of Correlational Research
Difference between correlational and quasi studies is not clear
- Correlation only uses DVs
- In a quasi experiment is you are using scores on dependent variable to create groups, based on demographics etc
- Correlation doesn’t split, it uses “continuous” variables instead, cannot use something that is a category, instead compare something that varies across a spectrum
Example of a Descriptive Study (People with Animal Bites and People Presenting with Depression)
- Asked How many people present with animal bites at emergency rooms
- Asked How many people presented with depression
Are these things related?
- 47.5% with cat bite had depression
- 71% with dog bites also had depression, and so on
Conclusions
- Suggest that there is a highly likelihood following a cat bite
Alternative explanations
- More likely to keep cats with depression
- Owning cats causes depression rather than bite
- Cats more likely to bite depressed people
- Therefore there is a big issue with causality
Descriptive Study (“Better Looking People are Happier”)
- Westernised demographic results suggest that “better looking people are happier”
- Self-report, quality of life questionnaire
- Interviewers rated attractiveness
- Justified by earning more money
- This cannot be a causal statement
JS Mill’s Criteria to Infer Causation
- Covariation
- Temporal Sequence
- Eliminate alternative explanations
Correlation is not Causation
- In informal logic, arguments try to suggest that two things relate because they co-occur
- Cannot say this, because you can’t isolate the direction (can’t control the variables) in correlational claims
How do we calculate correlations?
Have DVs
- Have to be continuous variables that have a scale, not categorical variables
- For each subject we are going to measure them on both DVs
Correlational Relationship
- The two variables covary in the same direction, in different directions, or do not covary at all
- As scores on one variable increase scores on the other variable increase, vice versa, or as one variable increases the scores on the other variable are unrelated
Positive Correlational Relationship
two variables co-vary in the same direction
- as scores on one variable increase, scores on the other variable increase and vice versa
Negative Correlational Relationship
two variables co-vary in different directions
- as scores on one variable increase, scores on the other variable decrease
No Correlational Relationship
two variables do not co-vary
- as scores on one variable increase, scores on the other variable are unrelated
Types of Correlational Relationships
- Direct - x–>y
- Indirect - x–>z –>y
- Spurious - x–> or y –>
- 3rd Variable - z–>x and z–>y
Examples of Some Correlational Findings
- Death by becoming tangled in bedsheets and cheese consumption
- More ice cream consumed, the more murders there are
Sources of Confounds in Correlational Designs
Person confounds - individual differences that tend to co-vary
- e.g depression and feeling lonely
Environmental confounds - situations that cause multiple differences
- e.g coming to UNSW can increase knowledge and anxiety (two things that are very different but are happening at the same time)
Methodological Sources of Confounds
Operational Confounds - measure that measures multiple things
- e.g correlation between impulsivity and poor decision making
- in fact the definition of impulsivity is poor decision making
Limits of Correlational Research
Correlational studies look at the relationship between measured variables
- Can establish covariation
- Cannot establish temporal sequence effectively
- Low in internal validity
- Cannot eliminate alternative explanations effectively
Confounds can arise due to
- Individual differences
- Environments
- Operational definitions