Methods of Empirical Research (Fach) / Einführung (Lektion)
In dieser Lektion befinden sich 8 Karteikarten
gaols, usw.
Diese Lektion wurde von witec_lerner erstellt.
Diese Lektion ist leider nicht zum lernen freigegeben.
- Goals of Empirical Research in Economics and Mgmt Developing theories and hypotheses regarding economic relationships Testing hypotheses regarding economic relationships Measuring the strength of effects in economic relationships Predicting economic developments
- empirical Research process Discovery of a problem -> Relevant? -> No ->End Yes -> State of Research -> Empirical investigation necessary? No ->End Yes -> Exploration -> Hypothesis generation -> Research design/operationalization/collection method/sample -> Pretest -> instruments proven -> no -> view at evaluation method again yes -> main collection -> (possible description) statistical testing of hapotheses with a suitable evaluation method -> interpretation
- Hypotheses - definition Scientific hypotheses are assumptions about structural properties of reality (Gabler Wirtschaftslexikon). They point beyond an individual situation and can be disproved by empirical data. Criteria for (scientific) hypotheses: general validity (universal proposition), i.e. going beyond an individual situation or event falsifiable, i.e. it must be possible to imagine events that are in conflict with the conditional clause. Possible formulations (amongst others): Conditional clause (if X – then Y, the A – the B) Equality/Inequality (mean of X = mean of Y)
- Information content The information content of hypotheses becomes higher when the IF component is logically extended and when the THEN component is limited: If a large pharmaceutical firm increases its R&D intensity from 5% to 10%, its profits will increase by at least 30% and not more than 50%.
- hypothesis: falsification Verification vs. falsification What we need: theories that are corroborated and ?correct so far?: an ideally large number of predictions derived from the theory were shown to not be wrong (i.e., not in contradiction to observations) – but for each of these predictions it would have been possible, in principle, to make observations that do contradict it "falsification" no observation so far has contradicted a prediction derived from the theory Such predictions are called hypotheses. Why falsification instead of verification? Of course, it is possible to make predictions about reality that are verifiable. E.g.: "There are black swans." Finding a single black swan will verify the prediction. However, before such a prediction has been verified it must be considered incorrect. Accordingly, also the theory itself would have to be considered "incorrect so far" or "incorrect until proven otherwise". And we don’t want to work with incorrect theories – they are at best "theory candidates". We develop hypotheses that are falsifiable. Only Hypotheses that have repeatedly not been rejected will be maintained as "proven hypotheses".
- Hypotheses: Falsificators A possible data combination which leads to the rejection of a hypothesis is called a falsificator. Returning to the issue of information content of hypotheses: The Information content of hypotheses increases with rising number of falsificators. Formulating hypotheses with low information content can be opportunistic – the researcher is protected from a possible rejection of a hypothesis. However, maintaining the hypothesis provides little gain in knowledge. Therefore: hypotheses should have the highest information content possible.
- types of hypotheses – Association hypotheses (A and B are related) – Difference hypotheses (A is different from B) – Causality hypotheses (A acts causally upon B): Causality does not yet follow from the finding of an association: – "When I write Christmas cards, Christmas is coming." → The association cannot be contested. But is there a causal link between Christmas cards and Christmas? Whether the presumption of a causal link is reasonable or not is, inter alia, a matter of study design. Experimental designs are most suitable to prove causal links. Associations usually have several possible causal explanations.
- Errors in statistical hypothesis testing In a statistical hypothesis test, the probability of an alpha error (error probability) is determined. If the latter is low, H0 can be rejected in favor of H1! h0 is actually true, h1 was said to be true based on the sample -> alpha error h1 is actually true, h0 was said to be true based on the sample -> beta error In empirical research, this statement would be checked as follows: Statement to be checked: Coin is unfair. – H0 (Null hypothesis): Coin is fair, i.e. probability(number) = 0.5. – H1 (Alternative hypothesis): Coin is unfair, i.e. probability(number) ≠0.5. Why so complicated? – The H1 hypothesis is (typically, such as here) formulated in a way that it can never be falsified. – Because "coin unfair " is also the case when probability(number) = 0.501. To falsify this, you would need about 1 million coin flips. That takes time. And even then it could still be the case that probability(number) = 0.5001 etc. – Therefore typically the approach to test and possibly reject the null hypothesis ("there is no association" or "the dice is fair," etc.). (The null hypothesis is an equality hypothesis!)