Research Skills 3 Flashcards
What are research ethics?
Refers to a written code of value principles that we use to make decisions about what is acceptable practice in any research project
What ethical code does psychology rely on?
BPS code of ethics and conduct (2009)
Code of human research ethics (2011)
What situations may result in ethical considerations?
Vulnerable groups (LD, Children, lacking capacity)
Sensitive topics (Ptps sexual behaviour, experience of violence)
Deception
Personal/sensitive and confidential information records
What is informed consent?
Involvement of research should be entered into voluntarily, knowingly and intelligently (Koocher & Keith-Spiegel, 1998)
(Israel & Hay, 2006, p. 61):
- researchers need to provide participants with information about the purpose, methods, demands, risks, inconveniences, discomforts, possible outcomes of the research, and how results are disseminated.
When might informed consent not be necessary?
Anonymous completion for questionnaires (although content may be sensitive and require prior information.)
Observational research in a public setting.
Types of deception?
Passive deception (withholding information):
- hypotheses
- purpose
- elements (e.g. priming)
Active deception - misleading participant:
What must be considered in deception before use, and what must be done if deception is used?
Depends on nature and seriousness:
Researchers should provide as full information as possible
Researchers must explain any deception at the earliest opportunity
Alternative approaches avoiding deception should be considered in cost-befit context
Ptps likely response needs to be considered
Points that contribute to the integrity of the researcher?
Whether the researcher has plagiarised
providing paper credit for contributors
Make raw data available for verification (storage)
Transparency
Bias in reporting
BPS guidelines for animal studies?
Avoid or minimise discomfort
Replace animals with non-sentience whenever possible
Minimise no. of animals used
When would you use an ANOVA (i.e. over a T test)?
Compare mean of continuous DV between more than two groups of a single categorical IV
Compare mean of continuous DV between two or more groups on more than one categorical IV
Only ONE continuous DV, have to have categorical IV.
What is GLM?
General linear model = particular way SPSS does ANOVAs, always means you are doing an ANOVA
Other names for the Dependent variable and the independent variable?
Dependent = Outcome
IV = Experimental or Factor
Fundamental principle of an ANOVA?
So we know the variability of the IV (because we’ve manipulated it) how much of this variability is accounts for the variability (results) of the dependent variable.
This is the F-ratio
SSr = residual = random variance that you can’t explain by you IV
SSm = model = Variance you can explain from you model
A highly significant result will have a larger amount of SSm.
Why is it advantageous to run one ANOVA rather than lots of T-tests?
All tests are probability based i.e. each time there is a 5%. If you do lots you may make a type 1 error (incorrectly reject null). One test (calculating an F-ratio) reduced this to only one 5% risk.
What are you interested in interpreting when you have one independent variable?
Only Main effects. NOT interaction.
Assumptions for a one way ANOVA?
Independent sampling (not influenced by other)
DV has an interval scale (distance between is equal)
Normally distributed within levels (DV)
Variance needs to be similar within each group. (homogeneity in independent groups, sphericity for repeated measures)
How can you check normal distribution (skewness and kurtosis) in you SPSS output?
Take the statistic and divide by Std Error, if it is above 1.96 then it is significantly skewed/kurtosed (p
What formal tests are there for normality in SPSS?
Shapiro wilk, Kolmogorov-Smirnov
If the test for homogeneity of variance in SPSS is significant what does this indicate?
They are not homogenous (bad)
Good if non-sig
Two ways to calculate effect size from 1-way ANOVA?
Independent groups (manual):
- eta squared (total variance explained)
- Omega squared (better- adjusts for random error)
Repeated measures (can request in SPSS) :
- Partial eta squared (variance uniquely explained by that variable)
- Omega squared
What are pair-wise comparisons? Problems (if not done as a test)?
Essentially lots of t tests between the different levels of the IV.
Family-wise error rate is high (lots of 5% unreliability) if you just did lots of t-test
How does a bonferroni post-hoc test work?
It does pairwise comparison tests but pushes the p value down to a more conservative value so the risk of type 1 error is not as high.
What are planned contrasts?
When we look at all possible comparisons we could make, decide that we are only interested in comparing.a few levels.
When reporting results of a 1-way ANOVA what is the layout?
F(df-model,df-error) = F-stat, p-stat, type-of-effect-size-test = effect-size-stat
Layout for reporting pairwise comparisons of a 1-way ANOVA
Mean difference [95% Confidence interval1, 95% CI2]
Layout for reporting planned contrasts of a 1-way ANOVA?
For each contrast (one row in SPSS)
t(df) = t-stat, p = p-value, r = effect-size
What situation is Tukey superior to Bonferroni?
Tukey is better when making LOTS of pairwise comparisons.
What situation calls for a factorial ANOVA (including 2-way)?
When there is more than one categorical IV. 2 IVs = 2 way, 3 IV equals 3 way and so on.
Fundamental principle in factorial ANOVAs?
Independent Groups: Like a 1 way but the variability for the model is split in 3. One section for IV 1, one for IV 2 and one for IV 1+2.
Repeated Measures: The same but you can discard another chunk that is between factors (so it should be more powerful)
Why do a factorial ANOVA and not lots of 1-way ANOVAs?
Family-wise error rate would increase
Can also look at interactions
Definition of an interaction?
The impact of any IV is conditional on a level within one or more of the other IVs
When analysing results of a factorial ANOVA who order to you look at results in terms of interactions and Main effects?
Check assumptions
Always look at highest order interactions first i.e.. 3 way then 2 way if doing a 3-way ANOVA
Then check main effects - but only if they are not negligible due to the above interactions
How can you tell on an interaction plot whether the main effects are confounded by the interaction?
So if you average the distance between the points you have on the y axis, and draw a line between them and you can see there is no overlap then there must have been a main effect irrespective of the interaction.
What can confidence intervals tell you about the interactions (and MEs) you have found (if you’ve found one)?
Interactions:
Looking on an interaction plot the confidence intervals can be used like error bars, to see whether the variability of the values will apply to the population. If the intervals overlap then the interaction is non-substantial.
You need to look at this in terms of all the other levels
MEs:
This also applies for MEs, if the CI of the MEs overlap then this is also non-substantial.
In a mixed ANOVA design what extra assumption do you need to check for?
Homogeneity of variance - Levene’s test on every group
If Sphericity test is significant, what do you do?
Read off another row - greenhouse geiser
In a 3 way ANOVA what 7 main results do you need to consider in the analysis?
3 Main Effects (A, B and C)
3 3x2 interactions (AxB, AxC, BxC)
1 1x3 way interaction (AxBxC)
Meaning of a 3 way interaction?
The 2 way interaction itself is different depending on a third factor in which the 2-way interaction is occurring.
How are 3 way interaction plots presented?
Two 2 way interaction plots produced, one depending on the third factor, and compare.
What are standard contrasts?
ways to explore an interaction. (3-way) Look at the specific interactions within the 3 way, look at the other levels.
What is a ANCOVA?
Advantages?
The difference lies in the F ratio underlying theory. It is basically saying that if we are aware of another covariant that we do not think interacts, then we can take the covariant’s chunk out of the F ratio of variance.
Will make it more powerful?
When might you use ANCOVA?
In quasi-experimental designs (people are not randomly assigned), you can use it to control for a factor e.g. measuring sleep deprivation on driving skill and controlling for driving experience. Used to protect internal validity.
When can you not use a covariant in an ANCOVA (what do you have to look out for (assumptions))?
- You have to make sure the covariant is having an effect on the DV WITHOUT being dependent on the IV. E.g. in the driving example, driving experience (IV) is NOT dependent on sleep deprivation.
- The Covariant must satisfy homogeneity of regression, this means the covariates effect on the DV must be constant (would suggest it is related to IV)
What effect can accounting for covariates in ANCOVAs have on internal and external validity?
Protects internal validity (the effect is likely to not be confounded)
Can harm external validity (the effect is less likely to be applicable to the general population) i.e. if you are comparing for lots of factors that situation is less likely to arise in the general population.
When interpreting an interaction plot how can you tell if there IS an interaction?
Hold constant X axis variable: this is when you look at the difference in between the points at level 1 (x axis IV) and compare that to the distance between the two points at level 2 (x axis IV). If they are going in the same direction then the interaction will be different in size, if they are going in other directions, the lines will cross.
Essentially if the lines are running parallel
When interpreting an interaction plot how can you tell if the interaction confounds the main effects? (2 way interaction)
You need to do this for both of your IVs.
This is when you take the average between the two points horizontally aligned with each other at each of the x axis levels and see if there is any overlap, overlap = confounded. This is for your X axis variable.
For your Y axis variable you do a similar thing but instead of horizontally aligned x axis points you draw a line in between the two point vertically joined up for both lines, and check for overlap.
If the average is higher then you will have a significant main effect but if they overlap then they are confounded by the interaction.
What is making up the ‘slice of pie’ in the SSm of a 2-way ANOVA?
IV 1, IV 2 and also the interaction between them.
Methods for gathering qualitative data?
Interview
Naturally occurring: i.e. Diaries
Observational: participant observational
Structured: open-ended questionnaires
Collaborative: Conceptual encounter, Memory work, role play.
Two theories concerning whether qualitative data is generated or collected?
Realist theory of knowledge
- Facts exist independent of the researcher
- Researchers job is to find those facts
- Collecting data
Relativist theory of knowledge
- What are considered to be facts are always mediated by human understanding
- Researcher and participant co-construct facts
Generating data
4 Methods for analysing qualitative data?
Discursive: Conversationalist, discourse analysis.
What we say to what audience and what facts are brought into debates and discussion.
Thematic: Identifying themes after a conversation has occurred
Structured:
Instrumental: Give it a theme before the research begun. E.g. a feminist interview and analysis.
4 decision points needed for thematic analysis?
Wide or narrow
Inductive or theoretical: i.e. bottom up or top down. Inductive is bottom up, taking all the coding from the material. Theoretical is top down, using presumptions to guide the interview.
Semantic or Latent: i.e. explicit or implicit. Semantic is explicit, using words and themes to generate analysis
Latent is Implicit, it is the process of interpreting words to produce an implied theme
Epistemological assumptions: realist or relativist (see earlier flashcard).
Features of Qualitative research as a human science? (3)
Reflexive
- Considers role of the researcher in generating and analysing material
- Considers the phenomena studied may already be a social product
Situated:
- Pays attention to the context in which data is collected
- Analyses data with this context in mind
- Privileges naturalistic settings
Inductive:
- Generation of small scale, local theories from observations
- Research questions NOT hypotheses
Evaluation of whether quantitative or qualitative is better?
Depends on the research matter
Although is commonly influenced by the current research paradigm of the subject.
This will influence assumptions and questions of researcher, unless they are brave enough to ask other questions.
Quick run-through of the process of the scientific method?
Theory
Research questions
Research design
Collect data
Analyse that data (descriptive and inferential)
How is qualitative data presented in terms of writing up? What is the general rule about the research design?
Written up as an interpretation of data, evidence is presented as examples from the data (often quotes)
Does not attempt to manipulate variables (non-experimental)
Four qualitative approaches in psychology?
Grounded theory:
- systematic, flexible approach to constructing and developing theories in grounded data.
Discourse analysis:
- Reality as socially constructed through language
Conversation analysis:
- Analyses talk as action, talk as interaction
IPA:
- Subjective exploration of the ptps experience.
Characteristics of Grounded theory?
Production of theory that is itself grounded in the data gathered:
Stages of analysis:
Codes:
- identify key points of the accounts gathered
Concepts:
- Collections of similar content (codes); grouping data
Categories:
- broad groups of similar concepts, used to generate a theory
Theory:
- Collection of explanations used to address the research questions
Overview of discourse analysis?
Approaches to the study on language. Identifies types of discourse within it or the means by which discourse is created.
Discourse means verbal exchanges between people or a system of ideas, metaphors and generated constructs. (talking basically)
Much of the analysis is based on research founded by theoretical or political agendas
Overview of conversation analysis?
Analysing the structure of verbal interaction.
Describing the structure, order and sequence patterns of interaction
i.e. the study of conversation itself
Overview of IPA?
Influenced by phenomenology and hermeneutics.
Phenomenology: the study of subjective experience
Hermeneutics: The interpretation of something (subjectively). IPA uses double hermeneutics because the ptp interprets an experience and then you interpret their interpretation.
Purpose of IPA?
To make sense of data in a meaningful way to address research questions.
Stages of IPA?
- Open coding
- Identification of conceptual themes
- Introducing a larger super-ordinate structure (clusters)
- Write up/Summary
When is a good time to use interviews?
When little is known about a subject
When a topic is complex/subjective/not easily -quantifiable
Generate Detailed and meticulous accounts, sensitive to the context
Explore in further depth quantitative data§
Types of interview?
Structured
Unstructured
Semi-structured
Biographic interview
Dilemma interview
Feminist interview
Free association narrative interview
Narrative interview
Things to consider when thinking about the person who performs interviews?
What type of interview will it become i.e. Advocate, co-user, evaluator?
Insider/outsider evaluation: People use multiple identities to establish Rapport Widdicombe (2015)
In IPA what type of questioning should you use?
Open, non-directive, exploratory and answerable questions.
Advantages of focus groups?
Widens range of responses activated forgotten details and can release inhibitions.
Can explore norms, be careful about power and conflict
Types of transcripts?
Jeffersonian: All the things said i.e. can include coughs and others.
Playscript: One after the other, whole words.
Ethical issues surrounding interviews?
Informed consent
Age of consent
Researcher safety
Data security
Anonymity
Who owns the interpretation, interviewer or interviewee.
How can you consider validity and reliability in qualitative research?
In quantitative research these rest on assumptions of objectivity and on researcher bias. In qualitative research you acknowledge context and role of researcher themselves.
Criticism of interview research?
Deletion of interviewer
Inadequate context and process information
Under-use of detailed transcription conventions
Assumed shared understanding of social science categories and roles
How to tackle generalisability in qualitative research?
Have to consider the research on a scale, from an individual level (unique context bound interaction) to applicable research at a population level (where it might apply across broad groups and contexts).
In what situations would you use what post-hoc test?
When all the assumptions are met:
- REGWQ, or Tukey (Bonferroni is a more conservative option)
With unequal sample sizes:
- Gabriel’s or Hochbergs GT2
unequal variances and sample sizes:
- Games-Howell