PS2021 stats exam Flashcards
(167 cards)
1
Q
What is Significance and how is it measured?
A
P value is used to denote significance
How surprising our data is
2
Q
What is the null hypothesis
A
Suggests no effects
The hypothesis that there is no significant difference between specified populations, any observed difference being due to sampling or experimental error
3
Q
What is the alternative hypothesis
A
Suggests there is the effect that was predicted
4
Q
What is a p value, and what does a high and low one mean?
A
P values denote significance.
A lower p value is evidence to reject the null hypothesis
A higher p value is evidence to fail to reject the null hypothesis
5
Q
What is the difference between independent and repeated designs?
A
Independent designs have different participants and the same participants are used in repeated.
6
Q
What is a factorial design?
A
When there is more than one independent variable
7
Q
What are the different types of factorial design and what do they consist of?
A
Factorial independent design
Factorial repeated design
Factorial mixed design –> both repeated and independent conditions??
8
Q
What are the assumptions of a parametric analysis?
A
- Independence of observations
- Interval or ratio level data
- Normally distributed data
- Homogeneity of variance
9
Q
What is meant by independence of observations?
A
One ppt cannot influence another’s data
10
Q
What is meant by interval or ratio level data
A
Data is scored along a continuum
11
Q
What is meant by homogeneity of variance?
A
The variance in the data set across groups/conditions should be roughly similar
12
Q
How to test normality? In independent and repeated
A
Look at the histogram, data should cluster around the mean –> avoid skewness & kurtosis
If its independent: evaluate normality within each condition
If it’s repeated: evaluate normality of difference scores
13
Q
How do you test homogeneity of variance
A
Levene’s statistic has to be used for independent design
The p value needs to be not significant for there to be homogeneity of variance
14
Q
How do you report homogeneity of variance in apa format
A
F(df1, df2) = Levene’s statistic ‘based on mean’, p
15
Q
How do you test normality on SPSS?
A
The Kolmogorov-Smirnov test
The p value needs to be not significant for normality
16
Q
How do you report normality in SPSS?
A
D(df) = statistic, p = significance
D and p in italics!
17
Q
What to do if Levene is not significant in an independent t test?
A
This means that the assumption has been met
Use the equal variances assumed row and continue with the parametric t test
18
Q
What happens if Levene’s is significant in an independent t test
A
The assumption has been violated
SPSS does a correction to the df to correct and adjust the analysis slightly, if you violate the assumptions it makes it more conservative and difficult to get a significant result because you’ve violated that assumption.
19
Q
What do you do if normality is skewed in certain conditions
A
Use a Mann Whitney U test
U = ‘Mann-Whitney U statistic’, z = ‘Z statistic’, p = ‘Asymp. Sig’
20
Q
Do you measure homogeneity of variance in a repeated design?
A
Homogeneity of variance is not relevant for repeated designs!
Parametric assumption: Independence of observations
o One data point should not influence another
BUT – repeated measures, so can’t avoid this!
o If my memory is bad it will be bad in both conditions
In repeated designs random variance is reduced
o No individual differences between condition
In repeated designs there is a different assumption
o Sphericity
21
Q
What do you do if normality is violated in a repeated t test?
A
Use a Wilcoxon test
(z = ‘Z statistic’, p = ‘Asymp. Sig’)
z and p in italics
22
Q
What is a one sample t test?
A
One sample t tests compare the data from a single sample of participants to a “reference value”
23
Q
What statistical test should you run when comparing more than two groups?
A
ANOVA
With each t test there is a 5% chance of saying there is a difference when there really isn’t a difference (p = .050)
ANOVA controls for all of that type 1 error within it, you just get 5% chance of error overall
24
Q
What is Type 1 error
A
Falsely rejecting the null hypothesis
25
What is family wise error?
Inflating chance of finding something
26
What is experimental variance?
o Variability between conditions
| o Due to experimental manipulation
27
What is random variance?
o Variability within conditions
o Due to measurement error
o Due to individual differences
o Due to unmeasured variables
28
What is the ANOVA statistic and what does it show
F ratio ≈ Experimental variance / Random variance
29
What does a larger F ratio show?
The larger the F ratio the greater the variance between the groups that is explained by experimental manipulation
Large F ratios = significant results
30
How to report a one-way independent ANOVA?
F(model df, error df) = F ratio, p = significance
F and p in italics!
31
What is model df?
Model df --> based on IV number of conditions minus one.
Note - it's labelled as whatever the IV is under 'tests of between-subjects effects'
32
What is error df?
Based on number of participants in each condition minus 1, summed together e.g., (10-1) + (10-1) + (10-1) + (10-1)
Note - it's labelled as Error under 'tests of between-subjects effects'
33
What are the different ways of breaking a main effect of a one-way independent ANOVA?
Bonferroni t test (old fashioned)
Planned contrasts
Post hoc contrasts
34
How do you break down the main effect of a one-way independent ANOVA using a Bonferroni t tests?
• Compare conditions with t test
o But we would need 6 t tests!!!
• Familywise error: 6 * 5% Type I Error
• Bonferroni correction:
o Change alpha level (criteria for significance)
o Bonferroni alpha = .050 / number of comparisons to be made
o Bonferroni alpha = .050 / 6
o Bonferroni alpha = .008
• Run all pairwise comparisons with t tests (see last week)
o BUT – t tests are only significant if p ≤ .008
35
Why do you have to do a Bonferroni correction and how do you do this?
A Bonferroni test compares conditions with a t test, but it would require many t tests!
This would lead to family wise error as there is a potential 5% Type 1 error for every comparison
For this reason you have to change the alpha level (criteria for significance) by the number of comparisons to be made
Then you run all pairwise comparisons with t tests
- But t tests are only significant to the new alpha level e.g., .05 divided by 6 comparisons = p < .008
The 5% error is shared across all the tests so it won't go above this error --> the new significance level much stricter
36
When would use planned contrasts?
When you have a one tailed hypotheses (the direction of any significant differences is specified)
37
When would use post hoc contrasts?
When you have a one tailed hypotheses (a difference is predicted, but with no direction)
38
What are the different types of planned comparisons?
Simple - Compares the effect of each condition to the control (defined as “reference” and “first”) condition e.g., 1 v s2, 1 vs 3, 1 vs 4
Repeated - Compares the effect of each condition to the next condition e.g., 1 vs 2, 2 vs 3, 3 vs 4
Deviation - Compares the effect of each condition (except the first – defined as “reference”) to the overall effect e.g., 2 vs 1234, 3 vs 1234, 4 vs 1234
Helmert - Compares the effect of each condition to the overall effect of all following conditions e.g., 4 vs 321, 3 vs 21, 2 vs 1
39
What is a polynomial planned contrast?
Trend analysis: looks at patterns in the data --> can only be used in very specific designs, cannot be used with every ANOVA
- Trend analysis is only appropriate for continuous IVs
40
If you are running an ANOVA and the levenes test is significant (assumption violated) what test should you use?
use a Kruskal Wallis test
41
If you are running an ANOVA and normality test is violated what test should you use?
use a Kruskal Wallis test
42
How do you write up a Kruskal Wallis test
(H = 'H statistic', p < 'p value')
43
Break
Break
44
What is an independent one way ANOVA?
When you compare more than two independent groups with only one independent variable
E.g., Time of day (morning, afternoon, night) and IQ scores
45
What is an independent two way ANOVA
When you compare more than two independent groups with two independent variables
46
How do you write up an independent ANOVA?
- Which type of ANOVA did you run and why?
- -> What IV? Independent or repeated? Number of levels? What is the dependent variable?
- How do you break down the main effect?
- -> Planned or post hoc tests? Which contrast and why? Hypothesis!
- Is the assumption of homogeneity of variance met?
--> Give statistic using APA standards and interpret the finding
F (df1, df2) = F ratio, p =significance
-Was the main effect significant?
--> Give stats using APA standards and interpret the finding
F (model df, error df) = F ratio, p =significance
- Interpret the main effect
--> Use graph, descriptives and planned/post hoc contrasts
Use all their in combination to explain where differences are
47
When should you use planned/post hoc contrasts
If the result is not significant, give the descriptives but don’t go any further. Only use the planned/post hoc contrasts if your main effect is significant.
48
What is the assumption of sphericity?
Sphericity refers to the condition where the variances of the differences between all combinations of related groups are equal.
o How consistent is the manipulation across conditions?
o E.g., if we give participants an extra bar of chocolate, is the change in happiness consistent? Or inconsistent?
o Want consistent changes across conditions!
Are the changes between a participant’s scores between conditions the same as each other – are they all going up or down by the same amount?
We want to see similar variability in change scores
49
Why can't Levene's be used for a repeated design?
Variance in each population should be equal. The Independence of observations assumption is violated in repeated designs
o Variances across repeated measures conditions are likely to be similar, because they come from the same people!
o So, will have less random variance as there are no individual differences between conditions
50
How is sphericity measured and when is it violated?
Using Mauchly's test of sphericity.
When Mauchly’s test of sphericity is not significant The assumption of having sphericity is met
--> this is because there’s no significant difference in the variability of the different scores
51
What do you do when sphericity is violated?
If assumption is violated...
SPSS provides a “correction”
o Various options available, but...
o “Greenhouse-Geisser” correction best
52
How does the Greenhouse-Geisser correction work?
o Calculation of variance and F ratio will remain the same
o Changes degrees of freedom: Reduced df means a bigger effect needed to achieve a significant finding
o Correction is proportional to the magnitude of violation
o SPSS will automatically give you corrections..
53
How do you write out Mauchly's test in APA format?
W = 'Mauchly's W', χ2 (df) = 'Chi-Square', p = significance
54
How do you report G-G corrected stats when sphericity is violated?
F (model df, error df) = 'F statistic', p = significance
These are all taken from the Greenhouse Geisser rows
55
How do you write up a repeated ANOVA?
- Which type of ANOVA did you run and why?
- -> What IV? Independent or repeated? Number of levels? What is the dependent variable?
- How do you break down the main effect?
- -> Planned or post hoc tests? Which contrast and why? Hypothesis!
- Is the assumption of sphericity met?
--> Give statistic using APA standards and interpret the finding
W = 'Mauchly's W', χ2 (df) = 'Chi-Square', p = significance
If significant, give G-G corrected statistic using APA standards and interpret
F (model df, error df) = 'F statistic', p = significance
If not significant, give 'sphericity assumed' stats using APA standards and interpret
F (model df, error df) = 'F statistic', p = significance
- Interpret the main effect
--> Use graph, descriptives and planned/post hoc contrasts
Use all their in combination to explain where differences are
56
What are the different types of one-way ANOVA's?
* One-way repeated (for more than 2 repeated conditions)
| * One-way independent (for more than 2 independent groups).
57
What is a factorial ANOVA?
This is an ANOVA that includes more than one independent variable
e.g., if IV1 is handedness (left, right, ambidextrous) and IV2 is film watched (sci-fi or romcom) and DV is travel sickness
This would be analysed using a 3 x 2 independent ANOVA
3 (left, right, ambidextrous) by 2 (sci-fi, romcom)
58
What type of IV's are in Factorial ANOVA's
IVs should be categorical (conditions, groups).
59
What are the different types of factorial ANOVA's?
* 2 IVs = Two-way ANOVA.
* 3 IVs = Three-way ANOVA.
You can have even more IVs, but don't bother!
60
Why run a factorial ANOVA instead of separate ANOVA's or t tests?
Multiple analyses like this increase chances of false positives
Adding a 2nd IV allows us to analyse whether the difference seen for one IV is dependent on the level of the other IV
There's also more power
o Adding factors helps us to reduce the error term, by accounting for variance that would otherwise be unexplained.
o Reducing the unexplained/random variance in cake.
61
What hypotheses do you need in a factorial ANOVA?
You need a hypothesis for each independent variable and one for the interaction effect prediction
62
What contrasts or comparisons do you use in a factorial ANOVA?
If the hypothesis is one-tailed use planned contrasts
If the hypothesis is two-tailed then use post hoc contrasts
63
How do you report the three main effects in APA format?
Main effect 1 = F (df, error df) = 'F stat', p < significance, n^2 = 'partial Eta squared'
Main effect 2 = F (df, error df) = 'F stat', p < significance, n^2 = 'partial Eta squared'
Interaction = F (df, error df) = 'F stat', p < significance, n^2 = 'partial Eta squared'
64
What do you do after you have a main effect for a factorial ANOVA?
Use the multiple comparisons to break down a main effect. Then report the mean and standard error to understand which condition is significantly higher/lower.
E.g., • There was a significant main effect of handedness on sickness score (F(2,24) = 17.09, p
65
What does an interaction effect show?
An interaction can tell you if the differences across one IV differ depending on the second IV.
66
How do you break down an interaction effect?
If one of the IV's only has two conditions, we could run independent measures t-tests looking at one of the IV's for each level of the other IV.
Report all the t tests at each level t(df) = 't statistic', p = significance. Bonferroni corrections need to be applied here because SPSS does not know how many comparisons are being made (divide .05 by the number of comparisons)
When reporting the interaction, you must also report descriptive statistics.
If both IVs have more than two levels run separate one-way ANOVAs and use post-hocs to break them down. Use bonferroni when needed.
67
How do you write up a factorial ANOVA?
Which type of ANOVA did you use to address the research question? What are your predictions? What are your IVs? How many IVs? How many levels? What is the DV?
Were assumptions met? Report stats, even if NS: Homogeneity of variance or sphericity
Is the first IV main effect significant? If yes, interpret where those differences occurred. If 2 conditions, only need means. If 3+ conditions, need further contrasts (planned or post hocs).
Is the second main effect significant? If yes, interpret where those differences occurred. If 2 conditions, only need means. If 3+ conditions, need further contrasts (planned or post hocs).
Is the interaction significant? If yes, interpret where those differences occurred. Take one IV(a), do the comparisons across the other IV(b) differ within each level of IV(a)?
Graphically represent any significant findings.
What is the answer to the research question? Refer back to your predictions: do the findings support or contradict?
68
What are the advantages of a factorial repeated ANOVA?
e. g., 3x2 repeated ANOVA
- 3 (before, during, after) by 2 (chocolate, fruit)
The advantages include typically requiring fewer participants than independent measures and it reduces the possible effects of individual differences that exist between groups.
69
What hypotheses do you need for a factorial repeated ANOVA?
e. g., 3x2 repeated ANOVA
- 3 (before, during, after) by 2 (chocolate, fruit)
A hypothesis for each IV e.g., in this case a prediction about time points and a prediction about cake type.
Lastly, you need a prediction for an interaction effect.
E.g., Happiness scores will be highest after eating chocolate cake (compared to fruit), but the differences will be smaller before and during.
70
How do you interpret main effects?
If the IV has two levels, it does not need to be broken down, just look at the means to see what is higher and lower.
If the IV has three levels it needs to be broken down using planned or post hoc contrasts depending on the hypothesis.
71
How do you report the main effect for different time points in an ANOVA?
Look under tests of within subjects contrasts.
Look at the linear rows, this will tell you if there is a significant linear trend across the time points.
Report it in APA F(df, df error) = 'F statistic', p = significance, n^2 = 'partial eta squared'
E.g., • There was a significant main effect for time point (F(2,18) = 84.10, p
72
Whats the difference between an independent ANOVA, repeated ANOVA and mixed ANOVA?
In an independent ANOVA, all the independent variables have different participants. No repetition.
In an repeated ANOVA, all independent variables have the same participants. They take part in every condition.
In a mixed ANOVA, there is one independent variable that has the same participants, and these are split up into independent groups for the other independent variable.
73
When is sphericity not needed?
When there are only two levels of each repeated IV.
Sphericity only applies when there are more than two levels,
74
What is a confounding variable?
• An extra variable that was not accounted for.
• It could introduce an alternative explanation of your findings...
o It could mean your IVs/manipulations were ineffective.
o It could mask the effects of your IVs/manipulations.
o It could lead to false positive findings.
• Increases variance in your dependent variable.
75
How do you deal with a confounding variable?
```
Good study design!
o Avoid them in the first place.
Random sampling/assignment to conditions.
o Or matching groups using controls.
Within-subjects/counter balancing.
o Although not a perfect solution.
```
76
What is a covariate?
Dealing with a confounding variable, statistically by adding a control or covariate to the model.
77
Can you use a covariate as a solution to a poorly designed study?
It is not a solution for a poorly designed study.
o Statistics shouldn’t be used to fix poorly designed studies.
o Using post-hoc covariates to try and find significance is considered as p-hacking!
o Covariates should be theoretically/conceptually motivated...
78
What are the two stages of interpreting an ANOCVA?
Stage one: is the covariate significant? Does the covariate explain a significant amount of variability in the DV?
Stage two: After taking the covariate into account, is there more experimental variability than random variability?
79
What hypotheses do you need for an ANCOVA?
You need one hypothesis for the covariate e.g., x will explain a significant amount of variance in y
80
What are estimated marginal means?
• EMMs are adjusted using the covariate
o Reflect the adjusted descriptives
• To interpret the ANCOVA output we need the EMMs
o NOT the unadjusted descriptive statistics
81
If you run an ANCOVA can you also run an ANOVA?
If you run an ANCOVA – you must not also run the ANOVA!!!
But interestingly, you would get very different findings.
82
Why do you get different results for the F ratio when you run an ANCOVA and an ANOVA?
* Because the random variance is reduced in an ANCOVA, we can be more confident of the differences we find.
* The covariate can explain away this “unexplained variance”, reducing the “within-subjects/error” variance.
83
What is an appropriate covariate?
* Continuous variables...
* E.g., IQ, number of hours.
* Binary Variables...
* E.g., Left or Right handed- scores must be 0 and 1.
Variables with 3+ categories cannot easily be analysed.
• Requires complex dummy variable modelling.
Covariables should be justified by research, not decided after the fact.
• Don’t use a covariate to fix a problem in your design.
Don’t include too many, usually just a couple at most.
84
What are the assumptions of an ANCOVA?
Independence of the covariate and the treatment effect
| Homogeneity of regression slopes
85
What the independence of the covariate and the treatment effect assumption
The covariate should not differ across groups. Run a one-way ANOVA to check that the covariate does not differ across groups/conditions
86
What is the homogeneity of regression slopes assumption?
The relationship between the dependent variable and the covariate should be fairly similar across groups/ conditions. Run the model to include a dependent variable x covariate interaction (and hope it is non-significant)
87
How do you report an ANCOVA?
Which type of ANOVA did you run and why?
What is the IV? Independent or repeated? Number of levels? What is the dependent variable? What is the covariate? How do you break down the main effect? Planned or post hoc tests? Which contrast and why? Hypothesis!
Is the assumption of homogeneity met?
- Give statistics using APA standards and interpret the finding
- F (df1, df2) = F ratio, p = significance
Was the covariate significant?
- Give stats using APA standards and interpret the finding.
- F (covariate df, error df) = F ratio, p = significance
Was the main effect significant?
- Give stats using APA standards and interpret the finding
- F (model df, error df) = F ratio, p = significance
Interpret the main effect
- Use table, EMMs and planned/ post hoc contrasts
- Use all three in combination to explain where differences are
88
What is a Type 1 error?
Claiming there is an effect, when no genuine effect exists
89
What is HARKing and a possible solution?
Developing hypotheses to match research findings.
Pre-registration: Hypotheses developed and submitted in advance.
90
What is low power and a possible solution?
Small samples and risk of Type 1 Error (false sig.)
Pre-registration: Power analysis included in pre-registration process.
91
What is p-hacking and a possible solution?
Running multiple analyses to find significant results.
Data sharing: Data sets and analysis tools shared openly.
92
What is publication bias and a possible solution?
Not significant research tends to not be published.
Pre-registration: Papers reviewed and prior to data collection.
93
What are the different types of pre-registration?
There are different types of pre-registration, you can have more formal and rigorous types of pre-registration right the way through to something more relaxed.
The most formal is registered reports (peer-reviewed).
Then preregistration via open access platforms (not peer-reviewed).
Lastly, the most informal pre-registration is private/ local preregistration (not peer-reviewed).
94
What can effect the size of your p?
Sample size can influence the size of your p.
o The larger the sample size, the more likely you are to find a significant p-value.
o This is because as sample size increases, variance decreases.
95
What is power analysis?
Power of a hypothesis test is the probability of detecting an effect, if there is a true and genuine effect exists.
o Probability a test will correctly reject a null hypothesis...
96
What are mixed methods in psychology?
Qualitative methods may be used alongside quantitative methods in mixed methodology research e.g.
o Quantitative methods may tell you if something had an impact or not
o Qualitative methods may help you understand why something had an impact
• This can help evaluate and improve interventions
97
What are the advantages and disadvantages of qualitative psychology?
Advantages:
- rich source of data
- often naturalistic
- hypothesis & theory generation
Disadvantages:
- difficult to test findings
- hard to generalise
- small samples
98
What are the advantages and disadvantages of quantitative psychology?
Advantages:
- testing theory & hypotheses
- generalise findings
- large samples
Disadvantages:
- data is not rich
- often not naturalistic
- tendency for confirmation bias
99
What is converging findings in mixed methods?
Converging is the idea of when you find the same or very similar findings that is converging evidence for a proposition. E.g., using both a qualitative and a quantitative study to find something.
100
What is diverging findings in mixed methods?
Diverging is when you have opposing findings e.g., finding something using a qualitative method but not seeing the same using a quantitative approach. This is also informative as it suggests that there is something in the methodology that allows us to uncover something.
101
What is complementary findings in mixed methods?
Complementary findings/evidence, which is similar to converging, but the findings do not necessarily show the same thing. By utilising different methodology, we can uncover a little more about what is going on underneath the surface.
102
What is partial vs fully mixed methods research?
Give examples.
Whether one or more aspect of the research process (e.g., objective, data/method, analysis, inference) are integrated. In partial mixing you have two separate strands of research and the mixing only occurs in the write-up of the discussion.
Partial mixing - focus groups and an experiment --> findings from both are integrated in discussion.
Full mixing - experiment with both quantitative and qualitative responses
103
What is concurrent vs sequential in mixed methods research?
Give examples.
Are both methods deployed together, or one after the other? In concurrent both are at the same time.
Concurrent - quantitative questionnaire & reflection log
Sequential - focus group and then experiment, or experiment and then focus group.
104
What is meant by equal status vs dominant mixed methods research?
Give examples.
Does one type of method take precedence in the research?
Equal status - questionnaire with approximately similar amounts of qualitative and quantitative responses
Dominant - Focus group with a short questionnaire at the end
- Questionnaire with mostly quantitative responses
105
What are the limitations of mixed methods
* Research more complex to carry out
* Need to know more about a variety of methods and how these can be mixed/complemented appropriately
* Time consuming
* More expensive
* Poor mixing may lead to ‘weak designs’
106
What is thematic analysis?
Foundational method, identifies themes which emerge from the data
107
What is grounded theory?
Identifies themes used to derive theories, data collection continues until emerging themes ‘exhausted’
108
What is discourse analysis?
Interested in the use of language, features of speech within data, how meaning is constructed
109
What is interpretative phenomenological analysis?
Understanding experience of particular group of people
110
What is content analysis?
Categorising information in the data, less qualitative
111
What are the potential pitfalls in thematic analysis?
1. Failure to analyse the data - Use extracts instead
2. Using the interview questions as the themes
3. Weak analysis where themes don’t make sense, overlap too much or fail to capture data
4. Mismatch between analytical claims and data - Themes ‘identified’ unsupported by actual data
112
What is a top down thematic analysis?
* Used when looking for something specific in the data
* Theory driven
* Start by developing provisional codes using theoretical frameworks, hypotheses, key variables which are applied to the data
* Less rich description of the data, but more detailed analysis if a specific aspect of it
113
What is bottom up thematic analysis?
* Used when not sure what to expect or look for in the data
* Data driven
* Start by allowing codes to emerge progressively, and work up into themes emerging from the data
* Provides a rich description of the entire data as you cast a wide net
114
What are the different steps involved in thematic analysis?
1. Familiarising yourself with the data --> transcribing data, reading and re-reading data and noting down ideas
2. Generating initial codes --> coding interesting features of the data in a systematic fashion across the entire data set, collating data relevant to each code.
3. Searching for themes --> Collating codes into potential themes and gathering all data relevant to each theme.
4. Reviewing themes --> Checking if the themes work in relation to the coded extracts and the entire data set.
5. Defining and naming themes --> Ongoing analysis to refine the specifics of each theme , generate a clear definition and name for each theme.
6. Producing the report --> Select vivid and compelling extract examples, relate back to the research question and literature.
115
What is the difference between inductive and deductive reasoning?
Inductive reasoning is bottom-up, you start with data and draw ideas to work out the bigger picture. It's often used in qualitative.
Deductive reasoning is top-down, you start with the big picture and develop specific ideas and collect data. It's often used in quantitative.
116
What is interpretative phenomenological analysis (IPA)?
Understanding experience of particular group of people.
```
IPA informed by three major areas of philosophy of knowledge (epistemology)
o Interpretative (I): Hermeneutics
o Phenomenology (P): Phenomenology
o Analysis (A): Idiography
```
117
What is the philosophical assumption of hermeneutics in IPA?
We are interpretative beings, focus on how people interpret the world
o Understand their subjective experience of a phenomenon, their perception
o ‘Double hermeneutic’: participant tries to understand their experience + then researcher tries to understand/interpret their account first you get their interpretation of it and then on top of that the researcher interprets their experience
o Not interested in describing participants experience, but try to interpret how the participant made sense of it
118
What is the philosophical assumption of phenomenology in IPA?
How people make sense of their experiences, understanding their ‘lived’ experience – appreciates diversity of human experience
o Descriptive: emphasis on exploring the ‘essence’ of these experiences; attempt to get pure understanding
• Bracket out prior knowledge biases
Phenomenology says that we should appreciate the diversity and shouldn’t necessarily try and generalise across masses of individuals.
119
What is the interpretative part of IPA?
It's concerned with sense-making; how they interpret their experience
o Impossible to bracket out prior knowledge/biases, instead use them
o What researcher brings to the analysis is integral
The descriptive type of phenomenology focuses on trying to bracket out the prior knowledge, whereas the interpretive type suggests we can’t bracket out our biases and instead we should use them to our advantage. We shouldn’t necessarily suspend our knowledge, there may be things in our existing knowledge base that can be helpful in interpreting the experiences of participants.
120
What is the philosophical assumption of Idiography in IPA?
The study of the individual
o Focus on detailed in-depth analysis of individual cases
o Commitment to understanding how phenomena are understood by a particular people, in particular contexts
o Integration (if at all) comes late in research process
121
What are the different stages of IPA analysis?
1. Familiarisation - read and re-read the data. Try to access the participant's world, attempt to engage with their experience.
2. Initial noting - develop initial codes. Phenomenological focus that describes participants lived experience. Make initial notes in the left margin.
3. Developing initial themes - look for themes within the data based on your codes. May use psychological terminology/ theory to help you. Record in right margin.
4. Structuring themes - search for connections between themes - relationships between themes, natural clusters with shared meaning, themes may fit within hierarchical relationships.
5. Move to next case (if any) - repeat stages 1-4 with your next participant.
6. Looking for patterns across cases to be integrated as a master table in your final report.
122
What is discursive psychology?
A type of discourse analysis, it requires the re-thinking traditional notions of language.
The focus is not on what people are saying, but rather how they are saying things and using language.
123
What is the difference between discursive and conventional psychology?
Conventional psychology views language as a window into a participants mind.
However, discursive psychology views language as social action. People use language to achieve something and language has an action orientation. To understand the action of language, the context in which words are spoken need to be appreciated.
Discursive psychology doesn’t focus on the cognition of language like conventional psychology (which assumes information is a window into their mind , it focuses on the action of language because it assumes that we’re always doing something with our language to construct the reality around us.
124
What can discursive psychology be used to examine?
* How people manage stakes within social interactions e.g., people may try to manage stakes of blame within a social interaction – how can they try to minimise their own accountability for something that went wrong.
* How people use interpretative repertoires when talking about something – these are patterns in the way people talk about things
Examples of research q's
o “All I have to do is pass”: a discursive analysis of student athlete's talk about prioritising sport to the detriment of education (Cosh & Tully, 2014)
o “Oh you don’t want asylum seekers, oh you’re just racist: A discursive analysis of discussions about whether it’s racist to oppose asylum seeking (Goodman & Burke, 2010)
125
What sot of data is used in discursive psychology?
o As natural as possible, ideally naturally occurring talk (e.g., radio discussions)
o Often not practical, so semi-structured interviews/focus groups may be used however this is not a natural setting
o Must have an interactional element
126
What are the steps in discursive psychology analysis?
1. Familiarisation - read and re-read the data. Engage with it naturally and consider what effect it had on you and how these actions were achieved.
2. Coding - select material (i.e., portions of the data) that is relevant to your research question (e.g., the discourse under consideration).
3. Analysis - involves many aspects that an analyst may go back and forth between. Examine how things are said, their context and the discursive devices used. Search for variability/consistency and identify interpretive repertoires and where they might contradict.
4. Writing - write up analysis of the findings with a selection of extracts foe the final report.
127
Name some discursive devices.
Absolute case formulations - painting something as definitive e.g., "always..." "everyone thinks..."
Disclaimer - anticipate potential alternative/ negative claims e.g., "i'm all for equality but..."
Concession - Acknowledge actual or potential counter-arguments to appear nice balanced e.g., "one downside could be... though..."
128
What are r values and what can they range from?
Pearson's r show the relationship between two variables.
They can range from -1 (perfect negative) to +1 (perfect positive).
A value of 0 shows no correlation.
129
What is a residual and how does this relate to a line of best fit?
A residual is the difference between the raw data point and best fit line.
A small residual indicates a more accurate model.
A line of best fit is when residuals are minimised.
If you have very little random variability, all the data points will be really close to the line of best fit.
130
What is the mantra of correlations?
Correlation does not imply causation!!
There may be other confounding variables that can explain the relationship.
131
How do you report a Pearson's correlation?
Look at the correlations table in the SPSS output.
| r = 'pearson correlation', p = 'significance'
132
How do you compare correlations?
You can compare the strength of different correlations using a z score.
(z = 'z score', p = 'significance')
133
What are manifest and latent variables?
The manifest variables are the actual items in your questionnaire and latent variables are the factors that you’ve extracted from them that you can go on and analyse.
134
How do you make a factor?
How do we know which manifest variables belong together? Which make a latent variable?
If things are highly correlated, it’s likely that one item has something to do with the other.
SPSS creates a massive correlation matrix by looking at all the items in the questionnaire. The closer the value is to 1 or -1, the stronger the correlation is.
135
What are the different assumptions of factor analysis?
Kaiser-Meyer-Olkin (KMO) - Are there enough participants to have a reliable solution?
Determinant - The variables should be correlated, but not too much as this suggests that every item is
136
What are the different assumptions of factor analysis?
Kaiser-Meyer-Olkin (KMO) - Are there enough participants to have a reliable solution?
Determinant - The variables should be correlated, but not too much as this suggests that every item is re-electing the same underlying thing. This is bad because the whole point is to identify different factors.
Bartlett's test of sphericity - This is the opposite, you need some correlations otherwise you cannot group them together. FA is only appropriate if variables correlate.
137
Why is it important for determinant vs sphericity to balance out?
Determinant - shouldn't have all variables correlated.
Sphericity - must be some correlations between variables.
We want to satisfy both these assumptions and have a balance between having enough correlations to be confident that we can extract some real latent variables but not so highly correlated that the entire questionnaire was just one big latent variable.
138
How do you run a reliability analysis?
You use Cronbach's alpha to assess if there is an acceptable level of internal consistency within a questionnaire.
It has to be above .7 to be an acceptable level of internal consistency.
139
When would you consider getting rid of an item in a questionnaire.
You check Cronbach's alpha if item deleted under item-total statistics.
This shows what would happen to reliability if you got rid of an item. If there’s a big jump you may consider rewording the item or getting rid of it.
140
In a reliability analysis, what is meant by corrected item-total correlation?
o Correlation between item and entire scale
| o If not reliable, the correlation will be low
141
In a reliability analysis, what is meant by Cronbach’s alpha if item deleted?
o What the alpha would be if the item were deleted
| o If not reliable, the alpha would increase (improve)
142
What variables are there in a multiple regression?
* Outcome variable: only a single continuous variable
| * Predictor variables: can include multiple variables
143
What type of predictor variables can you have in a multiple regression?
```
o Continuous predictor variables
o Binary (two group) predictor variables
```
144
What do zero order correlations show?
They are used like descriptive statistics in multiple regression write ups.
They ask whether there are significant correlations between each individual predictor variable and the outcome.
145
What are the different stages of analysis in a multiple regression?
Stage one - how good is the overall model.
Stage two - how good is each individual predictor.
146
How do you interpret the overall model of a multiple regression?
The model summary shows the percentage of variance in the outcome variable explained by predictors.
Then you use an ANOVA to assess whether the overall model (including all predictors together) any good.
E.g., The overall model , with all predictors together, is a significant predictor of *the outcome variable* (F (regression df, residual df) = 'F statistic', p = 'significance'), which explains x% of the variance in *outcome variable*
Then you interpret the individual predictors and report B, t and p in APA format. What is the direction of the effect for significant predictors?
Graph any significant predictors - bar graph for binary predictors (include error bars and label axes aand scatterplot for continuous predictors.
147
What does the adjusted r^2 show?
It shows the amount of variance in the outcome variable that is explained by predictors.
Report the adjusted R^2 as it shows a percentage of the variance explained.
148
How do you report individual predictors in APA format.
Look under coefficients in SPSS.
Report B, t and p in APA format.
The B coefficient shows you the slope.
E.g., *predictor variable* is a a significant predictor of *outcome variable* (B = 'B value', t = 't value', p = 'p value') with 1 unit of *predictor variable* leading to a 'b value' increase/decrease in *outcome variable*
149
What is the difference between a multiple regression and hierarchical regression?
The key difference between a multiple regression and a hierarchical regression is that a hierarchical regression allows you to control for some kind of confound/control variable.
150
What does Model/Block one in a hierarchical regression show?
How much variability in the outcome variable can be explained by the control variable? Is this significant?
151
What does Model/Block two in a hierarchical regression show?
After taking into account the confound, how much variability can be explained by the predictor variables? Is this significant? Tells us the unique contributions of the predictors.
152
What does the final model in a hierarchical regression show?
Taking together the control and predictor variables, is the overall model a significant predictor of the outcome variable?
153
How do you interpret models 1 & 2 in a hierarchical regression?
Look at model summary in SPSS.
The left shows the whole model at each stage. Model 1 shows the confound only and then model 2 shows the confound + predictors.
The right side shows the change in each model from the previous block in the model:
Model 1 shows how adding the confound predictor changes the model over and above having no predictor.
Model 2 shows how adding the predictor variables change the model over and above the confound variable.
154
How would you report the results of a hierarchical regression in APA format?
Firstly interpret model 1 - the confound variable
- *confound variable* explains *r square change*% of the variability in *outcome variable*
- *confound variable* is significantly/not better as a predictor of *outcome variable* than having no predictor at all (F (df1, df2) = 'F change', p = significance)
Then interpret model 2 - all the predictor variables together
- The predictors explain a further *r square change*% of the variability
- Assing the predictors, significantly improves the prediction of the model, over and above the confound (F (df1, df2) = 'F change', p = significance)
155
How do you interpret the overall hierarchical regression model?
Look under the adjusted r squared of model 2 to find the percentage of all the variability that is explained by every predictor.
Report it in APA, find the ANOVA table
The final model is significant F(regression df, residual df) = 'Regression F statistic', p = significance, explaining *adjusted r squared statistic under model summary* of the variability in the *outcome variable*
156
How do you write up a hierarchical regression hypothesis?
Hypotheses: two stages
1. Control variable will explain a significant amount of variance
2. After accounting for the control variance, give a prediction for each individual predictor variable
157
What are the assumptions of a multiple regression?
* No multicollinearity
* Distribution of residual values
* Homoscedasticity
* Outlier effects
158
What do assumptions tell us?
o How valid are the findings of our model?
| o Is our model biased?
159
What is multicollinearity?
When predictor variables are correlated with each other – this is bad!!!
This is bad because e.g., there’s a lot of shared variance between a person’s height and weight. This can become problematic in a multiple regression because if you’re trying to predict an outcome variable (age), using height and weight, you don’t know which is more important as they’re so highly correlated.
160
What is the assumptions of residuals?
* Residuals should be normally distributed
* Use a histogram to see the distribution of residuals for each participant
If its normally distributed then the regression model shows no bias in predicting participants scores.
161
What is the assumption of homoscedasticity?
When there is a similar variance of residuals across the variable continuum (high and low outcomes). The model is equally accurate across all scores.
We want points to be randomly distributed across the whole graph. If there is a funnelling effect seen on the right or left it suggests that there is a different amount of residual variability going across the continuum of outcome scores.
162
What is meant by heteroscedasticity?
When the variance of residuals differs across the variable continuum (high and low outcomes). The model is more accurate with some scores than others.
163
How is homoscedasticity related to homogeneity of variance?
Homogeneity is about the random variability within each of the conditions. We want the spread of scores to be roughly similar.
But, there are no groups in regression... --> instead of looking at groups of variability, it’s a continuum of variability – the variability in the residuals should be similar across the entire continuum.
164
What effect do outliers have on the line of best fit?
They can really shift the line of best fit. They affect the slope and intercept.
165
What is an outlier and how many is acceptable?
Standardised residual greater than ± 2 --> any participant with a residual over this is an outlier
Up to 5% of the sample being outliers is ok!
166
How do you assess multicollinearity?
Look at the correlations table, you want the correlations between predictor variables to be within -.9 and +.9
- You can report it like r = 'Pearsons correlation' p = significance
To meet the assumption, tolerance should be more than 0.2, the variance inflation factor (VIF) should be less than 10.
167
How do you write up a regression analysis?
End of lecture 19
