CRIM 220 Lecture Notes - Lecture 3: Statistical Conclusion Validity, Internal Validity, Observational Error
7 Jun 2018
School
Department
Course
Professor
Lecture 3, Week 3: Chapter 4 Tuesday, January 23rd
Crim 220 Chapter 4: General Issues in Research Design
Structuring Criminal Justice Inquiry
Example: What is the recidivism rate for juveniles released from custody?
•Definitions: be specific, give details. Be clear
•Time: how long do you track these persons for?
• How you state the problem will affect the outcome
Causation in the Social Sciences
• One of the key goals of social science is to explain why things are the way they are
• The general notion of causation is both simple and complex
• The cause of science is inherently probabilistic
Criteria for Causality
• Criteria for inferring idiographic causation and inferring cause in nomothetic mode
of explanation
Criteria for Causality (Shadish, Cook & Campbell, 2002)
Variables have to be correlated
Inverse relationship are also possible (one going up and the other one down)
• Example: Smoking is ‘relaxing’ so it reduces nicotine withdrawal
Problem: There aren’t any perfect correlational relationships (things of interest do not
always occur together)
Perfect correlation might indicate an error, if this occur one variable may be a function
of another
Temporal precedence (cause occurs before what we think is being affected)
• Example: Smoking precedes cancer
• Time order is often unclear (e.g. drug use and crime)
No third variable involved
• Spurious relationship
Idiographic Causation
Nomoethic Causation
1. How credible and believable it is?
(explanations must make sense)
2. Whether alternative explanations were
seriously considered and found wanting?
(all other possibilities have been
eliminated, so remaining explanation must
be truthful)
1. Two variables must vary together; must be
empirically correlated (how strong must
the relationship be in order to be causal?)
2. Cause must occur before the effect (time
order of variables may be unclear and
there can be exceptions)
3. Empirical correlation between cause and
effect is not due to some other factor (third
variable)
11
Probabilistic- causal reasoning that certain factors make outcomes more or
less likely to happen.
•Example: Smoking causes cancer (can never be actually proven, no experiment
done with human)
• No experimental studies to eliminate rival plausible explanations
Necessary and Sufficient Causes
• There are two types of causes: necessary and sufficient
•Necessary cause- a condition that must be present for the effect to happen
•Example: you can’t be convicted without being charged
•Sufficient cause- condition (if present) that more or less guarantees the effect
•Example: pleading guilty is a sufficient cause to being convicted
• The discovery of a cause has to be sufficient and necessary in order for it to
produce the best research outcome, however many of the cause found tend to be
probabilistic and partial
Validity and Causal Inference
• Scientists assess the truth of a cause by considering the threats to validity
• When we make a cause-and-effect statement, we are concerned with it’s validity—
whether it is true and valid
• Certain threats to the validity of our interference exist
• There are reasons why we might be incorrect in stating that some cause produces
some effect
Validity of Statements
• Whether statements about a given cause and effect are true (valid) or false (invalid)
• Different from validity measures
Subtypes of Validity of Statements
Ways of Designing a Research
Validity threats
Statistical conclusion validity- ability to
determine whether a change in the
suspected cause (IV) is statistically associated
with a change in the suspected effect (DV).
• Basing conclusion on a small number of
cases always find extremely strong
relationships (sample size is important)
•Opposite effect: covariation is present
when there is no cause and effect
relationship
Internal Validity- Between two variables
causal associations or are due to the effects
of some other variable.
• From nonrandom or systematic errors
• The effects of one or more other variables
•Example: drug users sentenced to
probation over prison recidivate less)
12
Validity- whether statements about cause or measures are well-founded and
correct (correspond accurately to the real world). The approximate truth of
an inference. Valid derived from ‘validus’ meaning strong.
Validity threats- possible reasons why we might be wrong when believing
some cause produced some effect.
