Research design

Quantitative methods with hypothesis testing are designed around the principles of critical rationalism (Karl Popper). This philosophical position states that theory can never be finally verified but only falsified.

Descriptive statistics can only describe the sample as a subset of the population. Inferential statistics draws conclusions about the population based on the sample.

Decision Tree ## Observational

Descriptive character, suitable for forming hypotheses (but not optimal for testing a hypothesis)

Cross Sectional Study (e.g. Survey)

no group comparison
e.g. “Quality of Life of Master Students”

Correlational Study

no group comparison
e.g. “Time spent on self-study and general quality of life correlate.”

Analytical study

group comparison
Identification and quantification of effects / verification of relationships
Not fully suitable for hypothesis testing

Example: Cohort Study

Multiple cohorts (group of people with comparable inital conditions) are observed. One cohort is exposed to certain factors. Cohorts are compared after

Smoking and Alzheimers

Cohort Study

Cohort Study

Case Control Study

Retrospective comparing cases with controls (self-reports?)

Case Control

Case Control

Other Forms

Cross-over design: during study treatment and control switches
- used in randomized and non-ramdomized (quasi)
Before / After comparison
Longitudinal studies:
- Trend study: different times with diff. samples
- Panel study: different times with same sample

Variance as basic concept

Primary variance:
- Impact of (experimental) factors on the change / variantion of the output
Secondary variance:
- Variation of output caused by nuisance variables
Error variance:
- caused by measurement errors and random processes

Variance

Variance

Experimental Design

Example: determine which variables have an influence on petrol consumption.

One-Factor-at-a-time (OFAT)

Full factorial designs

Fractional factorial designs

Quality criteria of experiments

Internal Validity

is essential condition for external validity.
Changes in dependent variable clearly attributable to changes in independent variables.
Increases with decreasing impact of nuisance variables (Quasi-Exp -> Experiment).

External validity

Results of experiment can be generalized to population
increases with increasing naturalness (laboratory -> field)
high internal validity can be problematic for external validity (labor vs. field)

Population validity

selection of representative sample

Situation validity

generalizable to situations deviating from experiment (ecological validity)

Construct validity

adequate operationalization of latent variable (construct)

Sampling

Lecture 05

Sample Size, Power, Effect Size

Lecture 06