Quasi experiemtal designs and field research
Download
Report
Transcript Quasi experiemtal designs and field research
Quasi experiemtal designs
and field research
Variations in classical experiments
Post test only designs
Used when pretesting might affect the
results
To assess this could use a 4 group design,
experimental pre-post
Experimental post
Control pre-post
Control post
Non-equivalent groups
Comparison group rather than control
group (difference is random assignment)
Use of matching
Widom’s study of child abuse and
criminality
Abused matched with a non-abused group
on gender, race, age, and SES
Problem with matching
Non-equivalent
Predicting parole risk
Use of LSI
3 jails, one (comparison) not informed of
LSI scores
In all groups, LSI predictive of further
recidivism
Jails who knew LSI scores released more
low risk offenders
Non-equivalent
Telephone service areas with caller ID
compared to those without services to
observe effects on obscene phone calls
Would be deceptive if there are
differences in obscene phone calls from
area to area—probably not the case
Less complaints in caller ID areas
Cohorts
Using a particular group that all begin at
the same time, compare to a group that
began at another time. Make assumption
of equivalence
i.e., compare graduate students who
started in 2003 with those who started in
2004
Police class, those sentenced to probation
in a particular month, etc.
Time series designs
Interrupted time series design
Take baseline data, make an intervention,
collect data
See p. 196
First example, general trend, not clear
that the intervention made a difference
Current downward trend in crime
Time series
Second pattern: random fluctuation
3rd pattern: immediate effect (in this
example, incapacitation)
4th pattern: more gradual effect
(deterrence?)
Interrupted time series with nonequivalent
comparison group
Time series
Time series design with switching
replications
If similar changes occur in DV in different
places at different times, corresponding to
when the intervention was introduced, it
makes it likely that the IV did affect the
DV
Field research
Best for topics that can best be
understood in their natural setting
Example: nonverbal behaviors
Relationship between environmental
design and crime requires observation of
the environment
Pedestrians before and after street lighting
was enhanced
Types of participation
Full participant (deception, may affect
what is going on, safety issues,
incompatibility)
Observing around the periphery of
criminal activity
Observer-as-participant: identifies self as
research, interacts with subjects
Police patrol studies (ride along)
Types of participation
Going native problem
Complete observer, not part of the action
in any way. May be unobtrusive, or might
identify oneself as a research, but no
interaction
Less able to ask questions
Observation
Observe, sometimes ask questions
Questions are often more spontaneous,
unstructured “informal conversational
interview”
Listening and probes
Gaining access
Formal organizations
Sponsor, letter, phone call, meeting
Gaining access to subcultures
Sponsor/informant
May be people working with criminals,
such as caseworkers, police, probation,
lawyers, private investigators, treatment
centers, ex-offenders, hangouts
Selecting subjects
Snowball sampling
Potential biases, i.e., only people who
have been caught or treated
Purposive sampling, sampling dimensions
Group, location, time, weather
Recording observations
Cameras to take photos
Video recordings
Tape recorders
Field notes—what we know, what we think
happened
Sketchy notes, expand later, write out
everything
Unstructured observation
Recording observations
Structured observations
Instruments to guide observations
Environmental surveys (BJA), may be used
to plan strategies
Other observations
Possibilities listed, recorded with details as
they happen
Combining with other data
Linking research methods
i.e., combining observational studies of
neighborhoods with surveys of resident
perceptions and crime statistics
Examples of studies
Shoplifters
How much? How much is identified?
Participant observation—participants
pretended to be shoppers
# shoppers who stole divided total # of
shoppers
Sampled days and times
Shoplifters
Everyone who entered the store was
counted
Systematic random sampling used to
select subjects
Subjects followed and observed
# thefts divided by the total observed
shoplifters
To determine detection rates, research
staff were used as confederate shoplifters
and some of them were observed by
research staff assigned to make
observations (double blind)
Reliability of observers could thus be
assessed
Reliability could then be used to adjust
shoplifting rates
Other studies
Seat belts—how many people where seat
belts?
Sampled time of day, type of road and
observation site, density of auto
ownership
Explicit instructions p. 307
Racial profiling
Problem with comparing race of drivers
stopped with race of distribution for a
resident population (i.e., more nonresidents may be stopped)
Estimate of drivers by observing race from
toll booths
Estimate of # of cars eligible to be
stopped
Profiling
Having observers drive 5 miles over speed
limit, count the number speeding and their
race
Lambeth’s study—blacks 13.5% of drivers,
15% of those speeding. 35% stopped
were black and 73% of those arrested
after the stop were black.
However, speeding not the only violation
Profiling
Lange study—examined only cases going
15 or more miles over the speed limit
Bars and violence: illustrates flexibility of
the method, changed study from one
examining why bars were violent (even at
violent bars, a low frequency behavior) to
what situations led to violence
Bars
Young working class, two groups in an
encounter where they are strangers
Crowded, no entertainment
High level of drunkenness (cheap drinks)
Over aggressive bouncers