Advanced Research Methods Flashcards
1
Q
Definition DAG
A
Directed Acyclic Graphs are graphical representations of the causal structure underlying a research question.
DAGs help to visualize the causal structure underlying a research question
2
Q
What do you need for a DAG?
A
- Prior knowledge of the subject
- Data on all relevant variables
3
Q
Path
A
Any route between exposure X and outcome Y
connection between exposure and outcome
4
Q
Causal path
A
Follows the direction of the arrows
5
Q
Backdoor Path
A
Does not follow the direction of the arrows
6
Q
Open paths
A
All paths are open, unless they collide somewhere on a path
7
Q
Closed paths
A
A path is closed if arrows collide in one variable on that path
8
Q
When is an open path blocked?
A
When adjusting for a variable
9
Q
What do we want to know from an causal inference?
A
We are not interested in the outcome per se, we are interested in the role of the treatment in achieving this outcome
10
Q
Definition causal effect
A
In an individual, a treatment has a causal effect if the outcome under treatment 1 would be different from the outcome under treatment 2
11
Q
Counterfactual outcome
A
Potential outcome that is not observed because the subject did not experience the treatment
12
Q
Individual causal effect cannot be observed unless..
A
Except under extremely strong and generally unreasonable assumptions
13
Q
When can a causal inference be determined?
A
Only when three identifiability conditions are met in a study
14
Q
The three identifiability conditions
A
Positivity
Consistency
Exchangeability
If all conditions are met the association between exposure and outcome is an unbiased estimate of a causal effect
15
Q
Positivity
A
Each individual has to have a positive probability of being assigned to each treatment arms
16
Q
Consistency
A
The treatment has to be well defined
17
Q
Exchangeability
A
- The individuals assigned to the different treatment arms have to be similar
- It does not matter who gets treatment A and who gets treatment B
18
Q
How to meet the exchangeability condition
A
- randomized rct
Individuals are randomly assigned to one of each treatment - Matching
For each individual who gets treatment A, there is an individual who gets treatment B - Stratification
Randomly select individuals from different subsets of the larger population. Almost impossible - Adjustment
Control for factors that influence the association between the treatment and outcome
19
Q
Confounder
A
An variable that effects X and Y
20
Q
Ethnography
A
The task is to document the culture, the perspectives and practices, of the people in the settings. The aim is to get inside the way each group of people sees the world
21
Q
Correlation
A
A statistical relationship between the treatment and outcome
22
Q
relative risk or risk ratio
A
the probability of an outcome in an exposed group to the probability of an outcome in an unexposed group
RR = 1 exposure does not affect outcome
RR < 1 the risk of the outcome is decreased by the exposure
RR > 1 the risk of the outcome is increased by the exposure
23
Q
Risk difference
A
The difference between the risk of an outcome in the exposed group and the unexposed group.
24
Q
Absolute risk increase
A
When the risk of an outcome is increased by the exposure
25
odds ratio
is a statistic that quantifies the strengts of the association between two events.
OR = 1 A and B are independent
OR > 1 A and B are associated correlation
OR < 1 A and B are negatively correlated
26
confounding
bias caused by common cause of exposure and outcome
You have to include and control/adjust the variable
27
collider
Variable where two arrows collide. The variable has to be excluded
28
Blocking
adjusting for a variable amongst a path. Blocking can be done by adjusting for any variable along a path.
29
Unblocking
Adjusting for a collider, unblocking a path by adjusting for an already blocked path
30
selection bias
if there is no equal chance of person a or person b becoming part of the sample
31
publication bias
positive findings are more likely to be published, which can skew the results that we see
32
mediator
explains the relation between the independent and the dependent variable. It explains how or why there is a relation
33
moderator
is a variable that effects the strength of the relation between the predictor and the criterion variable
34
self selection bias
when individuals volunteer to be in a treatment group. The sample in not random
35
recall bias
systematic error that occurs when participants do not remember previous events omit details
36
survival ship bias
when some of many of the observations are falling out of the sample which changes the composition of observations that are left
37
healthy user bias
people who take vitamins regularly are likely to be healthy
38
omitted variables bias
variables are neglected that may be important in the relationship
39
regression equation
example:
Weight = B0 + B1 x heigh
B0= constant
40
p <0,05
difference is statistically significant
41
the chance of finding a statistically significant result depends on
- sample size
- variation in population
42
testing
gives dichotomous result yes/no
43
estimating
size/strength of estimated effect
44
interpretation 95% CI
if the study was repeated, 95% of intervals would contain correct value
45
data ministry
adding too many variables without any theoretical justitification
46
multicollinearity
highly correlated explanatory variables
47
extrapolating beyond the data
regression results are only valid for populations similar to that of the study sample
48
non linearity
the assumption in regression analysis is that the association between the exposure and outcome is linear, but the association may be logarithmic
49
Unadjusted analysis
a researcher only focuses on bivariate association of two variables, for instance the outcome and exposure
50
adjusted analysis
more variables are included in the analysis
51
Logistic
observed outcome is dichotomous yes/no
X is linearly associated with the log of the odds of the outcome
Ln(p/(1-p)) = constant + X1 x something
52
odds formula
odds = (p : (1-p))
53
OLS
Observed outcome is continuous
X is linearly associated with the outcome
Y = constant + x X1 + x X2
54
continuous outcome example
weight or height
55
dichotomous outcome example
weight > 70kg yes or no
56
OLS can be used to
predict the outcome
direction and size of effect
57
Logistic regression can be used to
predict the probability of an outcome
direction of effect
58
Qualitative research
- Holism
- Open, how, what and why
- Flexible, naturalistic setting
- observation, interviews, documents
- words, description, interpretation
- researcher as instrument; involved
59
Quantitative research
- Reductionism
- To test hypothesis, to prove an assumption or causality, predict
- Closed question; associations
- Controlled or structured experiment
- structured observation, surveys, measurements, data
- tables, measure, calculation, statistical test
- detached, external instruments, tests, surveys
60
discourse analysis
- language use is important and should be the object of study
- language can be strategically used for all kinds of purposes
- the use of language can have consequences: it can shape how we think and how we behave
61
Hodges et al gives three forms of discourse analysis
Formal linguistic discourse
Empirical / conversation analysis
Critical discourse analysis
62
formal linguistic discourse
studying text to discover grammatical and linguistic rules
Sentences, structure and grammar
63
Empirical/conversation analysis
- studying talk in interaction to understand social practice
- also non verbal language
- for example non verbal language
64
critical discourse analysis
- studying macro discourses to understand the reproduction of power
- society level competition health care providers
- solidarity
65
Alvesson and Karreman levels of discourse analysis
micro
meso
grand
mega
66
Micro
- Detailed study of text itself without wanting to make broader claims beyond
- just text
- textual details
67
Meso
- studying language use to understand broader social practices
- overlaps with empirical/conversation
- Daily talk and meaning for social practice
68
Grand
- studying discourses that structure organizational reality
- level of university, ministry
69
Mega
- Studying universal discourses that structure human reality/ the way we view the world
- capitalism
70
Immersion
Immerse in the environment you study, become part of the group and try to understand them
71
Insider/emic perspective
you need to become part of that part of the society, feel what they feel and what they think
72
ethnography details:
ethnographic studies zoom in on daily practices, in order to understand these in context
73
Organizational Ethnography
- Understanding organizations as cultural entities
- Understanding the micro, going in depth through participant observations
74
Organizational Ethnography in healthcare
- care as organized practices
- bottom up / critical perspectives on care
- empowerment of minority voices
75
Ethnography in practice
Abduction
Sensitizing concepts
Theory field theory
76
Abduction
theory driven
77
Sensitizing concepts
Are concepts that you keep in mind while doing research. It helps to get closer and zoom in to a theoretical perspective.
78
Theory-field-theory
New insights from the field
79
Subtle realism Mays and Pope
(Criteria for qualitative research)
- epistemic position
- there is a reality that can be studied
- reliability, validity, generalizability
- triangulation, fair dealing, respondent validation, attention to negative cases, clear exposition of data collection
- importance of neutrality
80
Relativism Rolfe
(Criteria for qualitative research)
- Reality is multiple and socially constructed
- Open to challenge and depends on purpose
- No predetermined criteria, appraisal resides in the eyers of the beholder
- not neutral
81
Three different reasons for examining associations through quantative research
- description
- prediction
- causal inference
82
exchangeability through..
- DAGs
- Design Study
- Interpret results
- Draw conclusions
83
Goal of description
- to identify patterns in data
- obtain factual information
- not explaining patterns
- not drawing causal conclusions
84
Description Statistical Methods
Bivariate analysis
- Continuous outcome
- Mean, median, interquartile range (boxplot)
- OLS with one exposure variable
Dichotomous outcome variable
- Proportions, percentages, frequency
- Mean, median per category
- Logistic regression with one exposure
No adjustment, full associations
85
Description: Design and interpretation
Population data: eg election results
- Observations
- No uncertainty
Sample data:
- observations
- no uncertainty
- testing
86
Description: evaluation
Are the results interesting?
Starting point for further research?
87
Prediction Goal
- Predict the future
- If you know ABC what can you say about D
- Not to draw causal conclusion
88
Prediction Examples
If you have these symptoms, you will probably have this disease
If you have watched these films, you will probably like this film
89
Prediction Statistical Methods
Multivariate Regression Analysis
- Theory or data driven
- Difference between line and observations is ideally 0
- expand equation as far as possible so it explains ad much variation as possible
90
Prediction Interpretation
- Predicting outcome variable as accurate as possible
- Reducing uncertainty (error ideally 0)
- Interpretation of individual coefficients usually irrelevant
91
Prediction Evaluation
- how good is the data
- how well does the regression model fit the data
- how well does the regression model predict the outcome of interest
92
Causal Inference goal
Estimating causal effects
93
Causal Inference Design
RCT, 3 identifiability conditions
94
Causal inference Statistical Methods
DAG
bivariate regression
multivariate regression
depending on research question, adjustment
95
Causal inference interpretation
Individual associations relevant, focus on X
96
Causal inference Evaluation
Assumptions transparent; to what extent was bias avoided
97
Strengths of qualitative research
- Rich description of processes and experiences
- Knowledge construction and power relations
- Moving targets and phenomena in formation
98
two perspectives on observational studies
1. avoid causal language
2. emphasize causal language
99
avoid causal language
- bias cannot be avoided with certainty
- describe association
- emphasize that causality cannot be inferred
100
emphasize causal language
- be transparent about the real objective of the study
- design the study carefully, be transparent about assumptions
- acknowledge that bias cannot be ruled out
101
observational dimensions
- Space
- Actor
- Activity
- Object
- Act
- Event
- Time
- Goal
- Feeling
102
Space
where the researcher volunteered
layout of the place
103
Actor
people involved
Doctors, nurses, speech therapists etc
104
Activity
A set of related acts by several individuals. eg medical consult
105
Objects
physical things present
for example: protocols, MRI, medical journals, shampoo, braces
106
Act
single action by one individual
107
Event
Something out of the ordinary
108
Time
A sequence of events
109
Goal
The goal of the actors that are being observed
110
feeling
emotions felt and expressed
111
gap spotting
Researchers reviewed existing literature with the aim of spotting gaps in the literature and, based on that, formulated specific research questions
- Conservative way to think about science, building further on previous research
- takes a long time
112
P
The chance of something happening
The chance to roll 6 is, 1 in 6
113
Relative risk interpretation
twee groepen delen door elkaar
1- dat getal
"Women are 7% less likely to be referred than men"
114
Risk difference interpretation
group - group = for example 5.9
Women were 5,9% POINT less likely to be referred than men
115
Problematization
It means taking something that is commonly seen as good or natural, and turning it into something problematic
116
Confusion spotting
The main focus in this way of constructing research questions is to spot some kind of confusion in existing literature. Previous research on the topic exists, but available evidence is contradictory
117
Neglect spotting
Spotting something neglected in existing literature is the most common mode of constructing research questions in our sample. It tries to identify a topic or an area where no (good) research has been carried out.
o Overlooked, under researched, lack empirical support
118
Application spotting
It searches mainly for a shortage of a particular theory or perspective in a specific area of research.
