Summary
Summary book 2: experimental and quasi-experimental design - Shadish, Cook & Campbell (2002)
- Course
- Institution
- Book
English summary of chapters 1 through 6 and 8 with pictures.
[Show more]
Connected book
Book Title:
Author(s):
- Edition:
- ISBN:
- Edition:
6 reviews
By: thian • 4 year ago
By: nyamnuki • 4 year ago
By: dylan20 • 5 year ago
By: anomiepje • 5 year ago
By: annijr • 5 year ago
By: AniekHagen • 6 year ago
Seller
Follow
Machteld2
Reviews received
Content preview
Methods and techniques of evaluation researchBook 2: Experimental and quasi-experimental designs for
generalized causal inference.
Chapter 1: experiments and generalized causal inference.
Experiment = a test under controlled conditons ttat is made to demonstrate a known trutt,
examine tte validity of a typottesis or determine tte efcacy of sometting previously untried.
Causal relatonstips are usually easily recognized by people, suct as tte number of tours spent
studying is a cause of your test grades. However, ttese causes can be more complicated, suct as:
Low grades demoralizing reduced subsequent studying lower grades
In tte example above ttere is a reciprocal relatonstip, wtict refers to tte relatonstip between two
variables (low grades and not studying) ttat cause eact otter.
Cause
Cause = tte producer of an efect, result or consequence. Tte one (suct as a person, event or
conditon) ttat is responsible for an acton or a result.
Mackie (1974) introduces tte term INUS:
Insufcient but non-redundant part of an unnecessary but sufcient conditon.
Example: Ttere are several ways in wtict a fre can start, suct as matct tossed form a car, a ligttning
strike or a smoldering campfre. However, ttis is insufcient (I) because a matct cannot start a fre
wittout otter conditons (suct as oxygen and dry leaves). It is non-redundant (N) only if it adds
sometting fre-promotng ttat is uniquely diferent from wtat tte otter factors in tte constellaton
(suct as oxygen and dry leaves) contribute to startng a fre. ll of ttese causes are unnecessary (U),
because a forest fre can start even wten say a matct is not present. lso, none of ttem is sufcient
(S) to start tte fre alone. fer all, a matct must stay tot long enougt to start combuston and it
must contact sometting like dry leaves. So all factors are part of a sufcient conditon to start a fre in
combinaton witt tte full constellaton of factors.
Keep in mind ttat usually many factors are required for an efect to occur, but we rarely know all of
ttem and tow ttey relate to eact otter. Ttis is one reason ttat tte causal relatonstips are not
deterministc, but only increase tte probability ttat an efect will occur. lso, causal relatonstips are
context dependent, so tte generalizaton of experimental efects is always at issue.
Ttere are two types of causes:
Manipulable causes = dose of a medicine, tte amount of a welfare cteck and tte kind or
amount of psyctotterapy.
Ttese are ofen tte topic of interest in natural experiments.
Nonmanipulable causes = age, gender or raw genetc material.
Ttese can be researcted in analogue experiments, in wtict an agent ttat is similar to tte cause of
interest is manipulated. Example: you can’t ctange a person’s race, but you can ctemically induce
skin pigmentaton ctanges in volunteer individuals.
Efect
To beter understand wtat an efect is, it is important to know wtat a counterfactual is.
1
,Counterfactual = sometting ttat is contrary to fact.
In an experiment we observe wtat did tappen wten people received a treatment. Tte
counterfactual is knowledge of wtat would tave tappened to ttose same people if ttey
simultaneously tad not received treatment.
Tte counterfactual is not observable and consists of reasonable approximatonss
Effect = tte diference between wtat did tappen and wtat would tave tappened.
causal relatonstip exists of:
- Tte cause preceded tte efect
- Tte cause was related to tte efect
- We cannot fnd alternatve explanatons for tte efect otter ttan tte cause.
lso, it is important to remember ttat correlaton does not prove causaton. Ttis means ttat a simple
correlaton does not indicate wtict variable came frst and a correlaton does not rule out alternatve
explanatons for a relatonstip between two variables, suct as a ttird variable (confound variable).
Otter types are experiments ttat test wtetter a descriptve causal relatonstip varies in strengtt or
directon under test conditon vs. conditon B (tten tte conditon is a moderator variable ttat
explains tte conditons under wtict tte efect tolds). lso, some experiments add quanttatve or
qualitatve observatons of tte links in tte explanatory ctain (mediator variables).
2
,Notce ttat knowledge of tte efect of manipulable causes tells notting about tow and wty ttose
efects occur. Nor do experiments answer many otter questons relevant to tte real world, suct a
cause is distributed ttrougt society.
Tte strengtt of experiments is in describing tte consequences atributable to deliberately
varying in a treatment = causal descripton
Tte weakness of experiments is in clarifying tte mectanisms ttrougt wtict and tte
conditons under wtict ttat causal relatonstip tolds = causal explanaton
Causal explanaton is an important route to tte generalizaton of causal descriptons because it tells
us wtict features of tte causal relatonstips are essental to transfer to otter situatons.
Ttere are several types of causaton:
1. Descriptve causaton = simple bivariate relatonstips between molar treatments and molar
outcomes. Molar means a package ttat consists of many diferent parts.
Example: psyctotterapy decreases depression. Psyctotterapy consists of parts as verbal
interactons, setng ctaracteristcs, payment for services or tme constraints.
2. Explanatory descripton = breaks molar causes and efects into tteir molecular parts.
Example: tte verbal interactons and tte placebo features of psyctotterapy bott causes ctanges in
tte cognitve symptoms of depression, but ttat payment for services does not do so even ttougt it
is part of tte molar treatment package.
In tte end causal descriptons and causal explanatons are in delicate balance in experiments. Wtat
experiments do best is to improve causal descriptons, ttey do less well at explaining causal
relatonstips.
Modern descriptons of experiments
Experiment = a study in wtict an interventon is deliberately introduced to observe its efects.
Ttere are several types of experiments, suct as:
1. Randomized experiment = an experiment in wtict units are assigned to receive tte
treatment or an alternatve conditon by a random process suct as tte toss of a coin or a
table of random numbers.
Most important features are:
- Groups are probabilistcally similar to eact otter on tte average.
- Various treatments being contrasted (treatment vs. no treatment) are assigned to
experimental units.
3
, - Outcome diferences are due to treatment diferences, not to diferences between ttose
groups ttat already existed at tte start of tte study.
Kind of related is tte true experiment, in wtict ttere is also an independent variable ttat is
deliberately manipulated and a dependent variable is assessed.
2. Quasi-experiment = an experiment in wtict units are not assigned to conditons randomly.
Most important features are:
- ssignment to conditons is by means of self-selecton (units ctoose treatment for
ttemselves) or administratve selecton (teacters/tterapists/legislators decide wtict persons
stould get wtict treatment).
- Quasi-experimental design features usually create less compelling support for counterfactual
inferences. Ttis is because quasi-experimental control groups may difer from tte treatment
conditon in may systematc ways otter ttan tte presence of tte treatment, wtict can be
alternatve explanatons for tte observed efect.
- Tte cause is manipulable and occurs before tte efect is measured.
- Tte researcter tas to enumerate alternatve explanatons one by one. Ttis is closely related
to a falsifcatons logic by Popper (1959): confrmaton is logically difcult, so tte case is to
identfy a causal claim and tten to generate and examine plausible alternatve explanatons
ttat migtt falsify tte claim.
3. Natural experiment = study ttat contrasts a naturally occurring event suct as an
earttquake witt a comparison conditon. It is not really an experiment because tte cause
usually cannot be manipulated.
4. Correlatonal study = a study ttat simply observes tte size and directon of a relatonstip
among variables. It is usually synonymous witt non-experimental or observatonal study. Ttis
is wten a presumed cause and efect are identfed and measured but otter structural
features of experiments are missing. Ttere are no random assignment, pre- or postest or
otter design elements.
Experiments and tte generalizaton of causal connectons
Main issue: most experiments are tigtly localized, but tave general aspiratonss
Experiments usually tappen in a restricted range of setngs witt a partcular version of one type of
treatment ratter ttan a sample of all possible versions. Ttey exist out of several measures and tte
sample is ofen convenient. nd tte problem is ttis:
Tteorists and people wto read tte experimental study are ofen not interested in wtat tappened in
ttat partcular, past local study. Ratter, ttey usually aim to learn eitter about tte tteoretcal
constructs of interest. So, ttey want to connect experimental results to tteories witt broad
conceptual applicability, wtict requires generalizaton at tte linguistc level of constructs ratter ttan
at tte level of operatons ttat are used to represent ttese constructs in a given experiment.
Keep in mind: experiments ttat demonstrate limited generalizaton may be just as valuable as ttose
ttat demonstrate broad generalizaton.
Causal generalizaton = a type of deductve reasoning in wtict an accepted casual correlaton is
applied to a specifc. Ttis type of argument is commonly used to support a claim of explanaton.
Example: Oreo cookies make ctildren tungry tterefore, ttese otter of brand sandwict cookies will
make ctildren tungry.
Ttere are two types of causal generalizaton ttat are commonly researcted:
- Construct validity = inferences about tte constructs ttat researct operatons represent.
4
The benefits of buying summaries with Stuvia:
Guaranteed quality through customer reviews
Stuvia customers have reviewed more than 700,000 summaries. This how you know that you are buying the best documents.
Quick and easy check-out
You can quickly pay through credit card or Stuvia-credit for the summaries. There is no membership needed.
Focus on what matters
Your fellow students write the study notes themselves, which is why the documents are always reliable and up-to-date. This ensures you quickly get to the core!
Frequently asked questions
What do I get when I buy this document?
You get a PDF, available immediately after your purchase. The purchased document is accessible anytime, anywhere and indefinitely through your profile.
Satisfaction guarantee: how does it work?
Our satisfaction guarantee ensures that you always find a study document that suits you well. You fill out a form, and our customer service team takes care of the rest.
Who am I buying these notes from?
Stuvia is a marketplace, so you are not buying this document from us, but from seller Machteld2. Stuvia facilitates payment to the seller.
Will I be stuck with a subscription?
No, you only buy these notes for $4.89. You're not tied to anything after your purchase.
Can Stuvia be trusted?
4.6 stars on Google & Trustpilot (+1000 reviews)
62890 documents were sold in the last 30 days
Founded in 2010, the go-to place to buy study notes for 14 years now