RESEARCH METHODS
Lec 1: Quantitative Methods – Introduction
Ontology: What is reality?
Epistemology: What can I know?
Induction: first observe reality, then formulate theory (interpretivism)
Deduction: first formulate idea, then find out if it makes sense in reality (positivism + realism)
Criteria for quantitative research:
1. Reliability (If you would replicate research would this lead to similar outcome?)
2. Internal validity (Is the causal inference claimed in the research valid?)
3. External validity (Do the results hold in a different context?)
Cross-section research: compares variable in one moment in time
→ large group of people is surveyed, every question is a variable
→ correlation research because looks for correlation between variables (e.g. Are men taller than
women?)
+ high reliability: large sample
+ high external validity: representative sample, can be generalized to population
- lower internal validity: hard to make causal claims, spurious relation (→ add control variable to
test causality), reverse causality, endogeneity (meaning: something related to your Y variable that is
also related to your X variable, but you don’t know what → add strong theory or other research
design)
Longitudinal research: same people are surveyed on different moments in time
+ better internal validity than cross-sectional: find out what follows what (eliminate
endogeneity)
+ more data available
- internal validity: variables do not always vary consistently (problematic to exclude
endogeneity), spurious relation (→need control variable!)
- lower reliability: hard to find participants + collect data
- lower external validity: participants drop out, some types more frequently (→ hard to
generalise)
, Experimental design: capture causal mechanism
→ manipulation: researcher changes something in one group (treatment group) and not in another
(control group)
→ randomization: every participant has equal chance to be in treatment group or control group
→ groups only differ in one way: the stimulus (all other ways similar)
+ high internal validity: reversed causality excluded (→ manipulation), spurious correlation
excluded (→ randomization), understand cause – effect relationship
- lower external validity: hard to find representative sample of population (e.g. extreme bias
to students), experiments are done in an artificial environment
- lower reliability: low numbers of participants, high chances that if you replicate study you come
to different outcomes
Reliability Internal validity External validity
Cross-sectional Very good Challenging Very good
Longitudinal design Average Good Average
Experimental Challenging Very Good Challenging
Lec 2: Confidence Intervals
Inferential statistics: generalize from small sample to larger population
Population: total set of observations that can be made
Sample: subset of research units from the population
Parameter: any numerical quantity that characterizes a given population based on our sample (e.g.
age, gender…)
Sample statistics: characteristics of a sample which we use to make causal inferences on the
parameters (e.g. age, gender…)
→ sample statistics are known, parameters not
Confidence interval: indicates the range that might contain the true value of an unknown population
parameter
“With 95% confidence the sample mean is located within the interval ±1,96xSE”
Central limit theorem: if an infinite amount of samples are used from a population and if these
samples are sufficiently large (N>25 is usually satisfactory), the sampling distribution will be normally
distributed (mean=0 and standard deviation=1)
Lec 1: Quantitative Methods – Introduction
Ontology: What is reality?
Epistemology: What can I know?
Induction: first observe reality, then formulate theory (interpretivism)
Deduction: first formulate idea, then find out if it makes sense in reality (positivism + realism)
Criteria for quantitative research:
1. Reliability (If you would replicate research would this lead to similar outcome?)
2. Internal validity (Is the causal inference claimed in the research valid?)
3. External validity (Do the results hold in a different context?)
Cross-section research: compares variable in one moment in time
→ large group of people is surveyed, every question is a variable
→ correlation research because looks for correlation between variables (e.g. Are men taller than
women?)
+ high reliability: large sample
+ high external validity: representative sample, can be generalized to population
- lower internal validity: hard to make causal claims, spurious relation (→ add control variable to
test causality), reverse causality, endogeneity (meaning: something related to your Y variable that is
also related to your X variable, but you don’t know what → add strong theory or other research
design)
Longitudinal research: same people are surveyed on different moments in time
+ better internal validity than cross-sectional: find out what follows what (eliminate
endogeneity)
+ more data available
- internal validity: variables do not always vary consistently (problematic to exclude
endogeneity), spurious relation (→need control variable!)
- lower reliability: hard to find participants + collect data
- lower external validity: participants drop out, some types more frequently (→ hard to
generalise)
, Experimental design: capture causal mechanism
→ manipulation: researcher changes something in one group (treatment group) and not in another
(control group)
→ randomization: every participant has equal chance to be in treatment group or control group
→ groups only differ in one way: the stimulus (all other ways similar)
+ high internal validity: reversed causality excluded (→ manipulation), spurious correlation
excluded (→ randomization), understand cause – effect relationship
- lower external validity: hard to find representative sample of population (e.g. extreme bias
to students), experiments are done in an artificial environment
- lower reliability: low numbers of participants, high chances that if you replicate study you come
to different outcomes
Reliability Internal validity External validity
Cross-sectional Very good Challenging Very good
Longitudinal design Average Good Average
Experimental Challenging Very Good Challenging
Lec 2: Confidence Intervals
Inferential statistics: generalize from small sample to larger population
Population: total set of observations that can be made
Sample: subset of research units from the population
Parameter: any numerical quantity that characterizes a given population based on our sample (e.g.
age, gender…)
Sample statistics: characteristics of a sample which we use to make causal inferences on the
parameters (e.g. age, gender…)
→ sample statistics are known, parameters not
Confidence interval: indicates the range that might contain the true value of an unknown population
parameter
“With 95% confidence the sample mean is located within the interval ±1,96xSE”
Central limit theorem: if an infinite amount of samples are used from a population and if these
samples are sufficiently large (N>25 is usually satisfactory), the sampling distribution will be normally
distributed (mean=0 and standard deviation=1)
