Lecture 19: Thinking critically about causality Flashcards
wat is het verschil tussen een confounder en een collider
confounder -> construct A en B (confounder = dezelfde oorzaak)
collider <- construct A en B (collider = hetzelfde gevolg)
wanneer heb je een distorted association bij confounder
als je er NIET voor corrigeert
wanneer heb je distorted association bij collider
als je er WEL voor corrigeert
Deus ex Ha voorbeeld
men higher risk of dying faster than women. but there are as many men as women -> if it is 50/50, then no god exists. als er meer mannen dan vrouwen worden geboren, is er dus een soort compensatie, en dan bestaat god wel.
observations are pre-loaded with our assumptions. het is dus eigenlijk impossible om de wereld objectief te bekijken. wat is de oplossing hiervoor?
we need to be explicit in our assumptions!
wat is een quasi experiment
A quasi-experiment is a research design that shares some similarities with traditional experimental designs but lacks full control over the independent variable(s) due to ethical, practical, or other constraints. In a quasi-experiment, researchers study the impact of an independent variable on a dependent variable, but they do not have the same level of control as in a true experimental design.
Key characteristics of quasi-experiment:
No Random Assignment: Unlike true experiments, quasi-experiments do not involve random assignment of participants to different groups. Researchers typically use pre-existing groups or conditions based on some natural division, such as gender, age, or pre-existing conditions.
Pre-existing Groups: Quasi-experiments often rely on groups that already exist, like different schools, communities, or medical conditions. Researchers don’t manipulate or randomly assign individuals to these groups.
Limited Control: Researchers have limited control over the variables involved. They cannot manipulate the independent variable as they see fit, which is a hallmark of true experimental designs.
Causal Inference: While quasi-experiments can provide valuable insights into causality, they are often less conclusive than true experiments because of the lack of random assignment and the potential for confounding variables.
Practical and Ethical Constraints: Quasi-experiments are frequently used when conducting a true experiment is not possible or ethical. For example, it might be unethical to randomly assign people to a smoking group to study the effects of smoking on health, but researchers can compare smokers and non-smokers who have self-selected their behavior.
Despite their limitations, quasi-experiments can still provide valuable insights into cause-and-effect relationships in situations where conducting a true experiment is impractical or impossible. Researchers using quasi-experimental designs need to be particularly diligent in controlling for extraneous variables and selecting appropriate statistical methods to address potential biases.
Quasi-experiments rely on counterfactual inference, which can be estimated by observing the same unit over time or creating nonrandom control groups. However, these methods are not as reliable as randomized experiments, and critical thinking is necessary to evaluate the quality of the evidence.
OKE
3 conditions of John Stuart Mill
(a) the cause must precede the effect, (b) there must be a relationship between the cause and effect, and (c) there should be no other plausible explanation for the effect other than the cause.
wat is er met quasi experimenten en 3rd variables
To establish a causal relationship, quasi-experiments can be conducted, which allow for some control over third variables that could cause both the cause and effect. However, it is impossible to control for all third variables.
6 threats to validity of causal inference
history, maturation, selection, instrumentation, testing, and regression to the mean.
Popper’s logic suggests …
that alternative hypotheses should be tested to falsify the conclusions we wish to draw, but falsification can never be certain.
A more cautious version of Popper’s logic involves ….
repeated tests across different researchers, times, and places with different biases
wat is er met critical thinking in quasi experimenten
This chapter focuses on critical thinking about causation in quasiexperiments. The reason for this focus on causation is not that other kinds of critical thinking are unimportant in quasi-experiments. To the contrary, every bit of critical thinking that was described in the previous chapter for randomized experiments also has to be done in quasi-experiments, such as choosing good independent and dependent variables, identifying useful populations of participants and settings to study, ensuring that the assumptions of statistical tests are met, and thinking about ways in which the results might generalize. However, the quasi-experimenter also has one more task to do - the critical thinking that takes the place of random assignmenL
Multicollinearity can be a threat to the estimation of regression coefficients in a regression analysis, because:
- it causes the standard error of the b coefficients to increase, making the estimates of the b coefficients less trustworthy;
- it causes the value of the explained variance of the model to decrease;
- hit makes it difficult to determine the individual importance of the predictors.
Mike Pence: 2 out of every three smokers does not die from a smoking related illness and 9 out of ten smokers do not contract lung cancer
wat is mis met deze redenatie?
haalt causality met causation in de war.
hij zegt: geen causality smoking -> dying, op basis van associatie smoking - dying.
je moet het vergelijken met non-smokers, als die minder doodgaan is er misschien wel een relatie.
common misinterpretation: hij denkt dat het beiden sufficient en necessary moet zijn, maar dit is niet zo!!
Let them first discuss a more simple question, namely, the operations of body and brute
unintelligent matter; and try whether they can there form any idea of causation and necessity, except that of a constant
conjunction of objects, and subsequent inference of the mind from one to another
wat zegt Hume hier?
het enige wat je observeert is de witte bal en dan dat de andere gekleurde ballen gaan rollen. = conjunction of objects
all the rest (“the balls started rolling because the white one hit them) is an inference -> something you add from stuff happening in conjunction!
maar de vraag is… wanneer mag je deze inference maken?
oke
3 criteria John Stuart Mill
X causes Y if, and only if….
- Priority: change in X precedes Y
- Consistency: change in X varies systematically with change Y
- Exclusivity: there is no alternative explanation for the relationship
Problematic academic achievements, drug
abuse, pregnancy at a very young age related to low self-esteem.
If we create a stronger positive sense of
self-esteem, those other problems will
disappear by themselves
welke criteria zijn gemeet, en welke niet?
consistency wel
priority niet (wss, staat in ieder geval niet duidelijk)
exclusivity niet (er kan een derde variabele zijn, bv SES)
waar moet je naar kijken als je kijkt naar alternatieve explanations
naar dingen die gerelateerd zijn aan beiden variabelen
People with poor reading skills make more erroneous eye movements, go back to the beginning of the sentence more often (regression) and have more fixations per line
of text.
Abnormalities in eye movements (oculomotor skills) cause poorer reading skills
priority: nee, want het kan ook de andere kant op zijn (poorer reading skills -> abnormalities in eye movement)
exclusivity: nee, conflation happens. dus geen duidelijk onderscheidt tussen de variabelen (want wat is oculomotor skills, is een index van de eyemovements etc).
consistency: ja
ziekte in London: pellagra.
2 explanations: poor sanitation of poor quality of food. een researcher wilde kijken naar connectie tussen sanitation en disease: he selected 2 patients, maakte van feces etc. balletjes. wat liet hij hiermee zien
hij kreeg niks er van, geen ziekte.
liet consistency zien: geen consistency want ze kregen het niet. dus geen causal relationship! is gefalsificeerd.
wat was de eerste hypothese over de pellagra en wat was hier dus mee
slechte hygiene -> ziek
dit is een post hoc ergo propter hoc fallacy: omdat iets er voor komt, betekent niet meteen dat het de oorzaak is!
wat was er met keratin en haar gezondheid
inversion of causa and effect!
want ouder is minder keratine, maar meer keratine toevoegen betekent niet dat dit het probleem oplost.
weapons dont kill people, but people kill people.
wat is hiermee?
when do we call something a cause? do we only call something a cause if it happens only because of that?
no! het gaat niet om sufficient en necessary!
INUS condition=
insufficient
but
non-redundant
part of an
unnecessary
but
sufficient
condition
insufficient =
alleen die variabele is niet genoeg
match does not lead to fire without oxygen
non redundant=
substantially different from situation without that thing
zonder match aansteken is heel anders dan met een match aansteken
unnecessary=
niet genoeg, andere combinaties van dingen zijn ook mogelijk
zonlicht, dry grass and oxygen
sufficient =
combinatie van deze factoren is genoeg om het gevolg te krijgen
hoe relateer je INUS aan de gun problem
In research we try to compare observations with a….
good counterfactual!
- A perfect counterfactual….
knowledge of what would have happened to each participant if they had not undergone a certain manipulation.
If we compare that knowledge with what actually happened, we know what the effect of the manipulation is. That “perfect” variant is physically impossible.
Research into the relative contribution of CDR vs. affiliation in initiations. E.g. Existing groups: student association with heavy hazing, with mild hazing, without hazing
wat is hier lastig aan
groepen zijn selectief: ene groep is meer sensation seeking -> daardoor meer extreme hazing rituals.
sensation seekers in een groep en de rest in een andere. dan kijk je naar relatie sensation seeking en hoeveel mensen elkaar leuk vinden.
wat is er met de vliegtuigen en de bulletholes
we kijken alleen naar de vliegtuigen die terugkomen. waar bij het dus niet fatal was waar ze zijn geraakt. dus we moeten extra bescherming toevoegen op de plekken waar geen bulletholes zaten, daarvan weten we dat ze fatal zouden kunnen zijn.
waar is dit vliegtuig voorbeeld een vb van
the effect of attrition!!
(welke vliegtuigen zijn er uit gevallen???)
