# CRIM 220 Lecture Notes - Lecture 9: Regression Analysis, Causal Inference, Internal Validity

131 views5 pages

Chapter 9: Eliminating Rival Plausible Explanations: The Experiment

How can we take advantage of the existing structure of these data so that our inferences about what’s going

on are both reasonable and justifiable?

Three research approaches

1. Classic experiment

2. Quasi-experimentation

3. Case study analysis

The three share a common underlying logic involving eliminating rival plausible explanations to make

reasonable inferences about causes and other processes

Each differ in the degree that they emphasize manipulative control or analytic control

oManipulative Control: The active and intentional manipulation of the setting by the researcher in

order to manimize clarity of inference by ontrolling rival plausible explanations

Experimenter exerts control over every aspect of the setting

oAnalytic Control: One of the two general approaches to research that atttempts to make

inferencees about causes

Isolating Causes: The Controlled Experiment

Superstitious behavior comes from causal attributions we make that may make us feel better but are

unlikely to have any real causal effect

oLucky ring

Before we could say that one event or person caused some other effect, we had to demonstrate three criteria

1. Temporal Precedence: Show that the thing we think is a cause occurred prior to any changes or

differences that we think might have been produced by it

a. Causes always come before effects

2. Association: Changes in the putative cause co-varies in some reliable way with its alleged effect

a. The effect must occur often

b. Often: More frequently than you would expect on the basis of chance or coincidence alone

3. Elimination of Rival Plausible Explanations: Putative cause per se that is responsible for changes in the

dependent measure, rather than related variables, nuisance variables, artifacts, or any myriad other

potential causal agents that might have been present

The Terminology and Logic of Experimentation

Experimentation begins when we recognize, create a situation that includes the phenomenon of interest to

us

Does not need to exist in the real world as the lab affords us the ability to deal with possibilities or realities

Correlation does not necessarily equal causation

So many different variables intervene between cause and effect

Independent and Dependent Variables

Independent Variable or the Treatment Variable:

Dependent Variable or Outcome Variable:

Internal Validity

Internal Validity: The extent to which differences observed in the study can be attributed to the

experimental treatment itself, rather than to other factors

Possible Threats to Internal Validity

1. History: Specific events occurring between the first and second measurement in addition to the independent

variables

find more resources at oneclass.com

find more resources at oneclass.com

oWhat other events might have occurred between the pretest and post-test that could account for the

results?

2. Maturation: Processes within the research participants themselves that change as a function of time

oGrowing older, more tired, hungrier, so on

oBiological processes happen over time to cause change

3. Testing: The effects of taking a test on scores in the second testing

oHaving taken a test, you may become sensitized to the issue involved in a way that you wouldn’t

have been otherwise

oPretest Sensitization:

oPractice effects also threaten internal validity

oUnsure if one has improved purely because of the practice the pretest gave you or because of the

independent variable had imposed on you

Regression Toward the Mean

Different threat to internal validity is known as statistical regression or regression toward the mean

oStatistical Regression or Regression Toward the Mean: The propensity of extreme scores on the

first testing to score closer to the mean or average of the group on the second testing

This occurs because chance events are unlikely ever to stack up to precisely the same degree on two

successive occasions

The change is more illusory than real

Any measuring we do is subject to a certain amount of random error, no matter how many precautions we

take to minimize it

These positive and negative chance influences will be distributed equally across a group, so that average

scores will be a good indication of their true score

When chance events stack up to a positive or negative direction, statistical regression may occur

The tendency of extreme scores to move or regress closer to the mean on a subsequent testing is known as

statistical regression or regression toward the mean

oThreatens internal validity whenever a group is picked because of the extremity of their scores on

a pretest

Statistical regression

1. Operates to increase obtained pretest-post-test gain scores among low pretest scores since this groups

pretest scores are more likely to have been depressed by error

2. Operates to decrease obtained change scores among persons with high pretest scores since their pretest

scores are likely to have been inflated by error

3. Does not effect obtained change scores among scores at the center of the pretest distribution since the

group is likely to contain as many units whose pretest scores are inflated by error as units whose

pretest scores are deflated by it

Selection biases and instrumentation changes also threaten interval validity

oSelection:

oInstrumentation:

Controlling for Rival Plausible Explanations

Most common way of controlling rival plausible explanations is to incorporate a control group or

comparison group

Starts off the same as our first group and is treated identically except the control group doesn’t receive the

independent variable

This gives us the pretest/post-test control group design

Any change in the experimental group that wasn’t observed in the control group must have been a function

of the one element on which they differed

find more resources at oneclass.com

find more resources at oneclass.com