Research Methods⚗️ Flashcards
P value
Probability of finding effect in sample if no effect in population (unsystematic variation)
If probability less than 0.05 can reject null
CI
95% confidence interval around the mean difference
Range of scores likely to indicate the true population mean
T value
Calculation of p value based on calculation of t value
If mean difference increases then t value increases and p value decreases
Effect sizes
Indicate if differences are psychological significant (not just statistically)
Independent T test
Between participants
Difference between variables (continuous DV, categorical IV)
Independent t test df
Total sample size - 2
(Two means are used)
Data that can be freely varied and still get same descriptive statistics in the sample
Independent and dependent t test assumptions
Data approximately normally distributed (histogram clear skew, if not use non-parametric test)
No clear outliers (boxplot, points outside the box)
Spread of scores (variance) is relatively equal in both groups, error bars
Levene’s test affects accuracy of t test if not equal
Paired t test
Within participants
Difference between variables (continuous DV, categorical IV)
Paired t test df
Df=total sample size - 1
One mean used
Data that can be freely varied and still get same descriptive statistics in the sample
Levene’s test
Assuming variance
Not significant (p more than 0.05) no significant difference in variance in each group (assume equal variance)
Significant (p less than 0.05) significant difference in the variance in each group (do not assume equal variance)
Independent t and paired t reporting results
Group 1 were (significantly) better (mean, SD) than group 2 (mean,SD)
t (df) =T VALUE, p = SIG 2 TAILED VALUE
This suggests that group 1 were better than group 2
Cohen’s d
Interprets magnitude of an effect independent of the scale used
For both independent and paired t test
Use mean and SD for both groups/conditions
Larger value indicates more pronounced effect (can be negative if opposite direction)
Calculating effect sizes
After finding a significant effect in a null hypothesis
Calculate Cohen’s d using the Cohen’s d calculator
Compare value to levels of effect size and state magnitude of effect
Bigger effect sizes indicate more important effect
Pearson’s correlation
Relationship between two variables
Continuous IV and continuous DV
Scatterplot relationship conclusions
How much trend resembles a linear pattern
Variation in x explained by differences in y scores or vice versa
Relationship between x and y can be explained by z
Relationship is chance
Correlation coefficient
Measure of effect
Direction (positive or negative) and strength of relationship (the more it resembles a straight line, between 0 and 1)
Correlation value strengths
R= 0.01-0.39 weak R= 0.40-0.69 moderate R= 0.70-0.99 strong
Pearson’s correlation assumptions
Data normally distributed (histogram clear skew, if not use non-parametric test)
No clear outliers (boxplot, points outside the box)
Linearity (plot on scattergraph to check)
Pearson’s correlation df
Total sample size (N) -2
Pearson’s correlation shared variance
R squared
Variation in scores in one variable that can be explained by variation in the other variable
Stronger relationship = more overlap and shared variance
Pearson’s correlation reporting results
The findings show a (STRENGTH AND DIRECTION OF CORRELATION from TEST SCORE) between A and B
The relationship was (SIGNIFICANT OR INSIGNIFICANT),
r (df) = TEST SCORE, p= SIG 2 TAILED
This shows that A…
What happens if p is 0.000
You write p< .001
Chi squared test
Relationship between variables
Categorical IV and categorical DV (transformed from continuous)
Calculating chi squared test
Compare observed and expected values, bigger difference indicates larger Chi-squared value (more likely to reject null)
Chi squared assumptions
No more than quarter of cells should have expected value more than 5
No individual cell should have expected value more than 1
Numbers in each cell should be independent, categories are mutually exclusive
Cramer’s v effect sizes
Less than 0.10 trivial
- 10-0.30 small
- 30-0.50 medium
- 50+ large
Larger indicates more important relationship (chi squared)
Shared variance= v squared
Reporting cramer’s v for chi squared
Cramer’s v =
This is interpreted as a (small/medium/large) effect
( x100)= percentage of the variation in A explained by B
Report chi squared
In the sample, % of A and % of B did something
Chi squared test showed a (significant) relationship between two variables
xsquared (df, N=NUMBER IN SAMPLE)=CHI SQUARE VALUE, P=ASYMPTOTIC SIGNIFICANCE
This suggests that A is more likely to…
Non parametric tests
Not normally distributed (check with histograms)
Uses ranks (focus where score stands in relation to others)
Less powerful, median most appropriate
Parametric tests with their non parametric equivalents
Independent t -Mann Whitney
Paired t test-Wilcoxon
Pearson’s correlation- Spearman’s Rho
Mann Whitney
Alternative to independent t test,
Compare mean ranks of two independent groups (between)
Categorical and continuous
Ordered according to score (temporarily disregard group)
Rank scores in order, average ranks of duplicate scores
Reporting Mann Whitney and Wilcoxon
There was a (significant) difference between A (median) compared to B (median)
Mann Whitney U= MANN WHITNEY DATA, p=ASYMP 2
This suggests that…
WILCOXON- use Z instead of U
Z value negative value can be stated as positive
Wilcoxon signed rank
Alternative to paired t test
Compare mean ranks of participants who scored higher on (within)
Categorical and continuous
condition 1 to mean ranks of those who scored higher on condition 2
Spearman’s Rho
Alternative to Pearson’s correlation
Compare correlation using ranked scores
Continuous IV and continuous DV
Reporting spearman’s Rho
Spearman’s correlation showed that there is a (strong positive) relationship between A and B
Rs= CORRELATION COEFFICIENT MYATTRAC, P=SIG 2 TAILED
This suggests that…
Epistemology
Aim of research is to understand or gain knowledge about the world (patterns, behaviour, principles)
Approaches to research (philosophical)
Positivism- only one reality, uncover through observations
Post positivism-acknowledge need for falsification, use observation
Phenomenological- reality socially constructed, use dialogue to make sense of subjective experience
Constructionism- not one reality, socially/culturally produced through interaction
Relativism-not one reality, relative to historical, cultural and social context
Methodology
Strategy or approach to research
Qualitative or quantitative
Quantitative
More positivist, uncover one reality
Deductive- start with theory, specific hypotheses, see if reality makes sense
Analysis- constructs (thoughts, behaviour)of interest are coded to numbers
Qualitative
Relativist, not one reality, subjective experience
Inductive-observations from people, form more general theory. Aim to find general themes or viewpoints
Analysis-focus on content, emphasis on words not numbers
Qualitative methods
Direct data collection- interviews, focus groups, questionnaire
Indirect data collection-observations, analysis of online material
Qualitative criticisms
Subjective, influenced by personal bias
Does not represent population (but doesn’t aim to)
Cannot be replicated (does not aim to)
Not systematic
Qualitative research methods-interviews
STRUCTURED-predetermined questions in order, ask same questions to all to compares minimise bias, descriptive or exploratory
SEMI STRUCTURED- Flexible schedule of questions, few are set, open to new directions. Inductive approach, explanatory or exploratory
Conducting interviews good practice
Simple language, no jargon Prompt when necessary Rapport for comfort Active listening, nodding Body language, eye contact
Focus groups
Group dynamic, reflect on viewpoint. Range of perspectives
Avoid sensitive topics (may withhold)
Researcher as facilitator (steer it)
Qualitative research methods- observations
NATURALISTIC-everyday setting, unaware of observation. Can be archived data. No direct manipulation
PARTICIPANT OBSERVATION-researcher immersed in setting/activities
NON PARTICIPANT OBSERVATION-Researcher observes, no active part
Ethical and practical considerations of research methods
Consent-problem in naturalistic setting, knowledge can influence findings
Confidentiality/anonymity-how answers will be used
Debrief-a problem in naturalistic setting
Uncertain if have all data, if everything has been raised
Researcher/participant bias
Four main steps in data analysis (content/theme)
Anchor data to research question
Transcribe data- orthographic (verbatim) vs non orthographic (paralinguistics etc)
Initial reading of data- themes and how interviewer may have shaped data
Systematic data- driven by research question,cyclical refinement
Aim to find structure
Content analysis
Combines qualitative and quantitative
Extract common topics and themes, count how often they occur
Deductive (predefined areas) or inductive (emerges from data)
Content analysis steps
Familiarise with data Generate initial codes Search for higher order themes Review topics and themes Inter rater reliability check Count occurrences of each topic/ theme Write report
Codes for content analysis
Code should fit with data
Aim for as few codes as possible but still represent the data
Present only relevant info in write up
Thematic analysis
Extract common topics and themes
Use quotes to illustrate key aspects covered in wider themes
Distinction between major and minor themes
Grounded theory
Start with data and develops themes to generate theory (inductive)
Reflexivity
Important in qualitative analysis to be reliable and systematic
Aware of researcher’s influence on data
phenomenological analysis
Analysis of a subjective interpretation of a topic
df for everything
Independent t test -2
Paired t test -1
Pearsons correction -2
chi square Says it on the column