1. Descriptive
   • Simplify and organize data
2. Inferential
   • Go beyond simple description and help us make inferences about the data

Null hypothesis

• Null is from Latin: means "not any". The null hypothesis says there is no statistical difference between the population means. Inferential statistics are used to test the null hypothesis.

Type I Error (a)

• Made when Ho (the tested hypothesis) is wrongly rejected.

• Made when you fail to reject a false Ho.

When do you use naturalistic observation?

• When the question being asked concerns the natural flow of behaviour in natural settings. In this case low-constraint, observational based research is appropriate/necessary.

Problems with low-constraint research

• Generalization
• Case specific
• No causal relationships

• Just somewhat more constraint that naturalistic observation. Focuses on the subject's behaviour with little influence or constraint forced on the subject by either the researcher or the setting.

Constraints in case-studies

• Usually not carried out in the natural environment. Carried out in special settings where the experimenter is able to intervene somewhat. Case-study research usually focuses on individuals. Usually looks at limited classes of behaviour rather than the total context and natural flow of behaviours.

Use case-studies when:

• Beginning to investigate a new area
• Gaining familiarity with subjects before a major study
• Focusing on natural flow of behaviour
• A single subject is to be studied at a time
• Demonstrating the effectiveness of a new procedure

Case-studies can provide:

• Descriptions of events
• Identification of contingent relationships between variables
• The groundwork for more constrained research
• Observations to negate general propositions

• Generally, you can not choose your subjects. If you are a clinician you can not select your patients. So you must remember that you may not be getting a representative sample of the population. When making generalizations you must be sure your sample is representative of the population.

Flexibility vs. replicability

• One of the beauties of low-constraint research is its ability to handle modification to the study as events unfold. On the downside, this makes replicability (repeatability) very difficult, if not impossible.

The observer

• This is another limiting factor. No matter how diligent, the observer can not possibly see everything. Or appreciate everything the subject may be telling them, etc. Also, there is always the possibility that the observer will unconsciously influence the subject in some way.

Final warning

• Do not over generalize.
• Do not draw causal inferences.

Correlational research

• Often an extension of naturalistic observation studies. Variables are not usually manipulated. Usually a single group of subjects that is a sample of the population. But, the correlational research design always measures at least two distinct variables and plans for measuring the variables are designed before any observation is begun.
• Remember, you can not prove a theory, but you can disprove one.

Sampling

• Make sure youÕre obtaining a sample representative of the population to which you want to generalize. Also, ensure the relationship between the two variables is the same in all segments of the population under study.

Analysis of data
Must compute an index of the degree of relationship between the variables in the study. Use correlation coefficients.

• Pearson product-moment
• Spearman rank-order

Differential research

• Observe two or more groups that are differentiated on the basis of some pre-existing variable.
• Neither independent or dependent variables are manipulated.

Confounding variables and artifacts

• Confounded variables: vary together
  
• Artifacts: a failure of constraint

Why are differential studies higher constraint than correlational?

• Minimizes confounding and therefore strengthens conclusions drawn from the study. There is no such comparable control in correlational studies.

Problems with correlational and differential research

• A correlation (i.e., an observed relationship) does not imply causality. For example, if A and B are correlated, then you could have 1) A causes B; 2) B causes A; or 3) some other factor, C, causes both A and B. Abstractly, all are equally likely. However, in real life one or more of the possibilities may appear implausible.

How to generate a research hypothesis

1. Initial idea
2. Statement of the problem
   • Clearly state the expected relationship between variables
   • State the problem in the form of a question
   • Imply the possibility of an empirical test of the problem
3. Operational definitions
4. Research hypothesis

Testing the research hypothesis

1. Null hypothesis
   • A statement that there is no statistically significant difference (i.e., no difference beyond chance) between groups, categories, etc. at the population level.
2. Confounding variable hypothesis
   • Rejecting the null hypothesis, while necessary, is not sufficient to draw a causal inference. You must rule out factors other than the independent variable having any effect on the dependent variable (i.e., that there are no confounding variables responsible for any observed effect). Basically, you are ruling out other possibilities. Notice that unlike the statistical (null) hypothesis this hypothesis is not directly tested. Rather, each confounding variable hypothesis is ruled out by first anticipating potential confounding variables and reducing their impact on the study before it is even run or by subsequently carefully inspecting the research design and procedures.
3. Causal hypothesis
   • This takes us back to the research hypothesis. By ruling out the null hypothesis and the confounding variable hypotheses we offer support for the research hypothesis. This states that the independent variable has the predicted effect on the dependent variable. However, we can have confidence, but never certainty in the results of even a well run study.

Validity

1. Statistical validity
   • When statistical procedures are used to test the null hypothesis. Are the results are due to some systematic factor (ideally, the independent variable) or chance?
2. Construct validity
   • How well the study's results support the theory behind the research. To reduce threats at this level the researcher uses clearly stated definitions and carefully builds his hypothesis on well-validated constructs.
3. External validity
   • The degree to which we are able to generalize the results of a study to other subjects, conditions, times, and places.
4. Internal validity and confounding variables
   • The heart of the experimental design goal: demonstration of causality.

Some major confounding variables

• Maturation
• History
• Testing
• Instrumentation
• Selection
• Attrition
• Diffusion of treatment
• Sequencing effects