Polski revision
standard error
The standard error is the (hypothetical) standard deviation of the sample distribution estimate of how far the values of the sample means differ from the values of the true population means
dummy variable
dichotomous variable, 0 is the reference and 1 is the main interest: "y on average is coefficient lower/higher for 1 than for 0"
R squared value (not adjusted)
explained variance in percentage
content oriented discourse approach
focus is on interpretation of meaning rather than performative function of language
logged odds interpretation
for each 1 unit increase in x, the logged odds of y increases by coefficient
Hypothesis coding
positivist, deductive steps: codes predetermined by researcher and applied to qualitative data with specific goal of assessing a researcher generated hypothesis
snowball sampling
recruitment of participants based on word of mouth or referrals from other participants
interpreting moderation effect
- effect of being x on y is moderated by moderation variable -For each 1 point increase in *main effect* the *direction* effect of being *other main effect* is *increased or reduced* by 0.29.
Types of data in quantitative
-cross sectional - time series multi-level experimental
odds ratio interpretation
1 represents an even chance, 0.5 means you're 75 percent likely to win (it's only the P/(P-1) odds of 1 happening is coefficient times as big
Dealing with Accounting
1. Be well prepared and knowledgeable about your subject 2. Check other sources on the events you covered in the interview 3. Seek a diversity of perspectives on the subject and events covered by the interview TRIANGULATION!
unexplained residual
1- Rsquare, the variance the model cannot explain
Self-Reflexivity
1. Document reflections and assumptions throughout the process 2. Systematize reflections with a record of how one anticipates that position may impact research 3. Bring others into reflection process 4. Show reflexivity when publishing research
Steps of thematic analysis
1. Identify and familiarise with data 2. Identify codes (initial codes) 3. Find themes in data 4. Review themes 5. Review each theme - define and name (a codebook) 6. Document analysis - analyse resultant themes and draw inferences, document data.
Assuring credibility
1. Prolonged engagement 2. Persistent observation 3. Peer debriefing (Having a peer look through) 4. Negative case analysis (revising hypothesis with hindsight) 5. Progressive subjectivity (evaluating researchers own developing constructions of reality)
KEY QUESTIONS POSED BY QUALITATIVE RESEARCH METHODS
1. What is the role of the researcher? 2. Does it make a difference who is asking the questions? 3. Do a researcher's ideals and values affect what they study and should it?
critical considerations in case selection
1. access to the information 2. interpretivists engage in casing to tease out what something may be a case of 3. include preliminary considerations on the relationship between your case and wider cases 4. Avoid selecting on the dependent variable (don't choose cases that confirm your theory)
Interviews as method
1. to study unobserved events 2. to explore and understand the motivations of political actors 3. to understand the inner workings of important political processes 4. elicit rich and often new information for positivists it is testing ideas, for interpretivists it is making sense of what happened
interpretation of standard error
A large standard error relative to the value of a coefficient in a regression will suggest that estimate in the regression is not a very precise one - the sample mean diverges substantially from the true population mean
Law of Large Numbers
A principle stating that the larger the number of similar exposure units considered, the more closely the losses reported will equal the underlying probability of loss.
Heteroscedasticity
A regression in which the variances in y for the values of x are not equal. technically residual values depend on value of one independent variable (too small standard errors and sig.) detected in scatter plat and residuals. transform dependent variable or add control
Critical junctures
An important historical moment when political actors make critical choices, which shape institutions and future outcomes. have potential to disrupt path dependency and mechanisms that reproduce behaviours!
Reflexivity
Attempting awareness of our own positionality and our inherent assumptions about the effects these may have Self-reflexivity and methodological reflexivity
Dependability and confirmability
Audit trail: transparently describing the research steps taken from the start of the research project to development and reporting of findings
Guest Lecture 1: Looking at relationship of European policy, risk taking behaviour on Refugees decisionmaking
Based in Turkey, did online survey - Subjective risktaking is significant in moving onwards but not return aspiration - Gender is significant: men are more likely to stay in turkey (likely because they experience discrimination more) - Religious refugees are more likely to stay in turkey - If youre more likely to take risks, you are more likely to get out of turkey - Risk taking attitudes significantly affect their mobility aspirations
BLUE
Best Linear Unbiased Estimates
Risks of historical analysis; Solutions to those risks
Bias and selectivity; triangulation, source criticism, transparency Accounting; Evaluate purpose of documents, diversity of sources and views Lack of context/ information not always clearly known; Consult contemporary sources like journalistic accounts, interviews memoirs, leaks Reliability; identify gaps, summarise debates
Dependability
Concerned with stability of data over time: methodological changes expected but must be documented
credibility
Congruence between the constructed realities of respondents and the reconstructions of researchers
Qualitative quality criteria
Credibility Transferability Dependability Confirmability
Structural elements of a discourse
Definition of the problem: what does the discourse frame the problem to be? Ethos of governance/action: what normative principles are used to care about the problem Methods or modes: What framings are being used to legitimise this Subjects of governance: who is affected and can do something about this?
non-normality of errors
Errors are not normally distributed incorrect standard-errors and significance. transform dependent or add control variables
Critical framing analysis
How actors create categories to define what is normal or different Frames guide the ways participants perceive their social realities and represent these to themselves and others
Critical discourse analysis
How discourses justify or reproduce inequalities: aim to expose the discursive power struggle behind politics identifying the structure / exercise of power through language in the dominant / hegemonic discourse
confidence interval
If upper and lower bound are on the same side of zero, then the coefficient is significantly different from zero (both have to be positive or negative)
Interpretation of slope
If x rises 1 unit, then y will (rise/fall) by slope Also answer if negative or positive and the size
Inductive coding
Interpretivist approach, starts with data and involves Initial coding, focused coding (sorting initial codes into categories) and second cycle coding (recoding with categories)
Issues with interviewing
Interviewees don't always mean what they say: what to help or improve researcher Interview accounting: rewriting facts/framing them in your favour Memory is imperfect and can deceive
Britt Vande Walle: Political party think tanks
Looking into political party think tanks across the NL and BE and their autonomy from the political parties they belong to Collecting interviews and analysing these Uses qualitative comparative analysis, identifies different conditions and looks for similarities and differences.
Case studies in comparative research
Most similar systems design least similar systems design
Argumentative discourse analysis
Neutral descriptions of how discourses emerge
interpretation of p value
Null Hypothesis (coefficient = 0) means "this independent variable has no effect" So if (hypothetically) x has no effect on y in the population we would be very unlikely (p < 0.000) to find a sample of this size in which the coefficient is as different from 0.000 as it is (i.e. coeff) So, conclusion: our sample allows us to estimate or predict a significant effect of length on weight
OLS Regression
Ordinary least squares regression, tries to minimise the square distance of points to the line
Transferability
Potential for other researchers to apply a study to another research subject
Core materials of historical analysis
Primary archival material Relevant literature Personal record- memoirs, oral histories New supporting interviews (case study)
Credibility
Prolonged engagement Persistent observation Member check (Peer review from the people you interviewed) Triangulation (data triangulation, investigator triangulation, method triangulation)
Interpretivist qualitative
Reality is socially constructed Researcher part of the study Inductive approach: collect data, analyse data, develop explanations and theory
Methodological Reflexivity
Research methods are not neutral tools, evaluate purpose of methods, data collection etc.
Multicollinearity
Several independent variables are highly correlated with each other. This characteristic can result in difficulty in estimating separate or independent regression coefficients for the correlated variables and huge standard error Leave out or combine variables VIF score needs to be below 10
meaning of coefficients in interaction effect
Tells the effect of b1 if b2 equals zero, vice versa (you can't say that it's the independent effect!!!)
maximum likelihood estimation
The procedure of computing the score for all possible parameter values to identify the parameter value that confers the highest likelihood score
Central Limit Theorem
The theory that, as sample size increases, the distribution of sample means of size n, randomly selected, approaches a normal distribution.
Uses of the Case Study Method
Theory Building for qualitative Theory testing for quantitative
Transferability
Thick description: describing not just behaviour and experience but context so that behaviour and experiences become meaningful to an outsider
single case study
Typical case: understand causal pathways Extreme case: to explore issues where you do not know what is going on Deviant case: test limits of an existing theory Least likely: difficult test for your theory Most likely: easy test for your theory
Spuriousness
When other factors (often referred to as Z factors) are actually causing two variables (X and Y) to occur at the same time; it may appear as if X causes Y, when in fact they are both being caused by other Z factor(s) Detected by thinking, then adding omitted variables as control variables
Reciprocity
X affects Y but Y also affects X, leads to incorrect coefficients. solved through time series analysis
Coding
a code is a word or phrase that captures an important attribute of your data (meaning that is relevant to the RQ)
spurious relationship
a false association between two variables that is actually due to the effect of some third variable
Process tracing
a method of connecting events in sequence to identify cause and effect; drawing descriptive and causal inferences from pieces of evidence
Logistic Regression
a nonlinear regression model that relates a set of explanatory variables to a dichotomous dependent variable
Challenges of Elite Interviews
access: locating and contacting interviewee Preparation Obtaining proper answers to questions Issues with accounting
Atypical observations
an outlier with extreme x value has leverage. this has strong influence on data. can be seen by rule of thumb: more than three SD away
historical analysis
analytical process of reconstructing historical events: sequencing of time and role of individuals and their decision-making Use this reconstruction to evaluate explanations and build test specific theories
Counterfactual in historical analysis
asking what if things happened a different way: helps look into historical factors that affected key decision making
second cycle coding
axial coding: identifying relationships among open codes Theoretical/selective coding: choosing or discovering the central or core category around the research story
Predicted probabilities
can only calculate for specific values of x! chance of 1 happening if x is given value
Key tools of historical analysis
causal impetus: consistency of interpretations with data and sustaining generalisable claims Consideration of alternative explanations: Demonstrate serious consideration of alternatives and use the counterfactual
Triangulation in interviews
corroborating the interview with other primary and secondary material
Types of Research Designs
cross-sectional, comparative, case study, mixed, longitudinal, experimental
Testing spurious relationship with multiple regression
does the effect decrease? does it stay significant?
Linear regression
estimation technique aiming to predict the value of a dependent variable y with the values of one or more independent variable x
coefficient
evaluate based on size and direction.
Christian Pipal: Text analysis of parliament and politicians speeches
generally, politicians that are not in power have an angrier tone. tone is not dependent on gender really. you can use AI, dictionaries, trained coders and crowdfunding tools to do this
Positionality
how the researcher's background affects the entire research process: Motivation, relationship with subjects and values/predispositions affect research
Foucaldian discourse analysis
identifying the existence of distinct discourses that shape what can be said / how problem is identified
interpretation of intercept
if a person's body height is 0cm, that means the persons weight is -54.8 (EVALUATE IF THIS IS MEANINGFUL)
Case study
intensive study of a single group, incident, or community, sometimes referred to as Small N analysis For the purpose of understanding a large class or group of cases
Reflexivity
keeping a diary of ones own experiences, explicit and implicit assumptions, preconceptions and values
Non-linearity
linear regression cannot test for non-linear! you can see this through scatterplots. Transform variable or do different types of regression
replacement of y in the logistics equation
ln( P/(1-P)) because it's a ratio! how likely is p to happen given that not p could happen
Values of Pseudo R square
measures the improvement in the likelihood of the model between the coefficients fitted as a result of a maximum likelihood calculation (iteration 5) and the likelihood of the model producing outcomes with both coefficients set to zero (iteration 0)
Thematic analysis
method for systematically identifying, organising and offering insight into patterns of meaning (themes) across data sets
principle for comparability in comparative research: Equivalence
need equivalent data, same linguistics, similar context, similar methodology
principle for comparability in comparative research: controls
need to control some differences and similarities between units of study
P-value
needs to be less than 0.05 Indicates the probability that sample value was obtained by chance- if h0 is true
Positivist quantitative
objective reality Neutral and replaceable researcher Deductive approach (Theory, Hypotheses and Test on data)
comparative qualitative research
often spatial and generally studies two units
Pseudo R squared
probability of finding exactly this sample if coefficients are correct/ likelihood of the model being perfect if the coefficients were exactly what the model predicts
Y axis of logistic regression
probability of y
Confirmability
process of assuring that data, interpretation and outcomes of research are not simply figments of researcher's imagination. data can be traced back
issues with purposive sampling
prone to selection bias Often challenging to show transferability of research results More risk that selected participants are too homogenous to describe the demographic
most similar systems design
researcher compares very similar systems in an attempt to explain differences between them
least similar systems design
researcher selects very different cases with similar outcomes
Non-independent errors
residual error of one observation should not tell you anything about the residual of another observation (data is nested within groups) too low standard error. possibly do time series analysis or multilevel
theoretical sampling
selecting sample members based on earlier interviews that suggest that particular types of participants will help researchers better understand the research topic
purposive sampling
selecting sample members to study because they possess attributes important to understanding the research topic
Experiments in qualitative
study in which the researchers retain 'control' over the recruitment, assign subjects to random conditions, introduce a treatment and compare results between groups
t value
t= coefficient/standard error (+-1.96)
equation for Pseudo R2
the Zero coefficient likelihood - Maximum likelihood divided by zero coefficient likelihood, all divided by 100 to produce a value between 0 and 1
warren's rule
the broader the scope of the study and the more comparisons the research intends to make between groups the larger the sample required. balancing act between not having enough data for saturation and having too much for in-depth analysis
proving spuriousness or mediation
the inclusion of the control variable reduces the size of the effect of the IV, possibly see a reduction or loss in the significance of other IVs
Research design
the set of choices we make about collecting and analysing data
discourse analysis
ways in which a discourse develops, is contested and becomes prescriptive or dominant What is at stake in an issue, how are problems defined and what assumptions underlie different points of view
mediation
when a third variable explains for the relationship between x and y
Partial mediation
when the significance falls slightly but not above 0.05
interaction effect
whether the effect of the original independent variable depends on the level of another independent variable (Multiply x1*x2 in equation)