Deck 6 Flashcards
1
Q
What is causality in Hume’s view?
A
Contiguity, temporal precedence and regularity of association - all that is needed to infer causality for Hume.”I keep seeing X first, then Y, thus I conclude that X causes Y”.
2
Q
Problems with Hume’s perspective on causality?
A
- It leaves no room for probabilistic causality, where X increases or decreases the chance or value of Y, but does not deterministically trigger Y.
- Temporal vs. Causal order. E.g. the rooster crows before the sun comes up, but that does not mean that crowing causes the sun rise.
- Confounding/ spurious relationships. Hume’s rules can not make a distinction between actual causal effects and spurious relationships on confounding.
3
Q
What is new and improved Humean perspective on causality?
A
- Associations may be probabilistic.
- Logical connection (plausible theory they for the causal connections)
- Non- spurious relation (elimination of competing explanation)
All of it is based on logic and theory. But establishing a plausible connection is not enough, as it may be easy to come up with plausible stories for the opposite explanations. The general take away is that many competing plausible stories about how the would works can be proposed and they don’t solve the problem of confounding. Statistics also cannot solve the problem, because simply cannot distinguish between several very different underlying structures. Consider the paths A -> B -> C and AC, statistically adjusting for B would break the link between A and C. However, in the first case, adjustment would make no sense, since we block a mechanism through which A exerts a causal influence on C, while in the second case, adjusting for B makes a lot of sense, because it is a confounder and biases our estimate of the causal effect of A on C when not adjusted for (or otherwise accounted for).
4
Q
How to break confounding/ biasing paths?
A
- The idea behind experiments is that actually intervening/ manipulating instead of merely observing breaks any confounding/ biasing paths because they, by definition, no longer play a role in influencing X we manipulate X ourselves under controlled conditions.
5
Q
What is Randomization?
A
- Randomization is a technique for assigning units to treatment conditions in experiments, theoretically both eliminate confounder bias by severing every incoming link to the randomized variable including the ones we don’t know about or cannot measure, and it enables the researcher to quantify uncertainty.
6
Q
What are the problems with consequential manipulation?
A
- countervailing effects (simultaneous mechanisms that compensate for one another).
- over - determination - if there are two causes of X, where only one is sufficient to trigger Y, experimentally manipulating any of them without manipulating the other one as well would lead to the mistaken conclusion that none of them are true causes of Y.
- pre - emption.
- experiments often struggle with external validity e.g. they intervene in natural states and the way they do so often leads to problems with generalizing findings to other more natural conditions.
- not everything can on should be manipulated.
7
Q
What is essential for making causal claims?
A
Causal claims need causal models and causal models need a good understanding of the concepts, the processes and the context (it is not a statistical exercise).
8
Q
What is sampling?
A
- Sampling is the selection of research units from a larger target population.
9
Q
Why do we need sampling?
A
- You want to draw conclusions for your target population but you cannot study all of it so you select a subset that represents your target population.
10
Q
What is a sampling plan/ steps in sampling?
A
- define the target population/ area/ stuff e.g. enterprises
- define an operational population e.g. all tax taxpaying enterprises
- define a frame - ideally it would match operational population.
- choose a method: random on non - random sampling
- decide on your sample size.
11
Q
What is random sampling?
A
- All elements in the population have a non - zero and known chance of being selected. The types are: simple random, systematic, stratified random, cluster, single- or multistage. The purpose of it is to get external validity by statistics.
12
Q
What is non - random sampling?
A
- Selection (partly) based on the judgement of the interviewer or researcher. The types are: quota, accidental/ convenience, judgmental/ purposive, snowball. The purpose of it is to get external validity by theory.
13
Q
What is simple random sampling?
A
- E.g. lottery system. Every element of the population has the same probability of being selected, you can achieve a simple random sample by numbering all elements in your population, putting the numbers in a fishbowl and drawing numbers from the fishbowl until you have enough elements for your sample. Or you can use a random table on a random number generator.
14
Q
What is systematic sampling?
A
- In systematic sampling you select a random number on a list map and then every other n-th element from the list. Watch out with this kind of sampling, take care of periodicity e.g. sampling newspaper every 7th day. But sometimes it can increase precision eg. By having a jump, there is less chance of selecting highly correlated units.
15
Q
What is stratified random sampling?
A
- Study population divided in sub -groups (strata). Random sample of research units are taken from each group, either proportionally or disproportionately. Strata are groups of units that possess specific characteristics that are analytically relevant (e.ge gender * age * rural/ urban). Strata comes from theory.
