Research Methods II.2

research hypothesis - an empirically testable statement derived from theory. It is stated in terms of the population. It holds for the whole population, but we only have a sample.

null hypothesis - the opposite of the research hypothesis given as a fixed value.

Assume the null - to derive a sampling distribution. A mathematical probability distribution, we do not find it empirically

Determine the critical region - 5% or less

Calculate a test (sample) statistic

Make a decision - infer about research hypothesis

One tailed test: a statistical test where extreme results leading to rejection of the null hypothesis can be located at only one tail

Two Tailed Test: A statistical test where extreme results leading to rejection of null hypothesis will be located at both left and right tails

• Type 1 Error: Rejecting a True Null
• Type 2 Error: Affirming a False Null

Probability of a type 1 error

The probability that the value of a parameter falls within a specified range of values. Constructed with Z distributions.

As sample size increases sampling distribution decreases and vice versa. This occurs because there is a smaller chance of an error.

How much better can we estimate Y, the dependent variable, using X, than if we estimate Y by itself

1. Lambda

2. Gamma

3. Kendall's Tau

4. "r squared"

5. "r" not PRE

For nonspurious bivariate relationships.

Usually involves the introduction of other variables to determine the links between the independent and dependent variables, or specification of the conditions under which association takes place.

1. Nothing (Z has no effect)

2. The correlation between X and Y reduces drastically. This may mean: spuriousness OR Z is an intervening variable

3. The relationship between X and Y changes but does not disappear: then Z is a conditional variable

(PRE) measure of association for two nominal variables or 1 nominal+1 ordinal variable. Problem: it equals zero when the modes all fall in one value of the dependent variable. Then you use a percentage comparison instead.

(PRE) measure of association for two ordinal variables. Can be negative, varies from -1.00 to +1.00; Problem: it inflates the correlation because it doesn't include “ties"

A measure of association for two ordinal variables. It corrects the gamma problem by including “ties.” NOT a PRE

It is a PRE, scale of 0 to 1, for intervals and ratios

Uses a scale of -1 to 1, both assume linear relationship, for intervals and ratios

• Demographics
• Time
• Place
• Interest
• Concern

What do you want to know? What do you expect respondents to know?

Question Content: Fact or Opinion, wording, provide middle alternatives

Question Type: Closed vs Open-ended

Response Format for Closed-ended

Question order

1. Topic: is it something people know about?

2. Screen for non-attitudes – Did the organization give respondents middle alternatives?

3. Question wording: Did the organization give you the question wordings?

4. Question Order: Questions can be placed to lead respondents in certain ways

5. Sample definition: Probability sample? From what population?

6. Sub-sample sizes and error margins. What happens if certain demographics are pulled out? (error margin increases)

7. Who sponsored the poll?

8. What's the spin?

Higher response rate

Can probe and clarify

Quick turnaround

High costs

Possible interviewer bias

No anonymity

Low cost

No interviewer bias

Greater anonymity

Considered responses

Lower Response Rate

Requires simple questions and answers

Nobody there to clarify/probe

Takes time (opportunities for outside phenomena to interfere)

Less expensive

May be the only source

You can replicate/triangulate someone else's research with your own

May only approximate the data you want

May have hidden biases

Sometimes access is a problem

Less prep time

Less expensive

Don’t need to construct a survey

With a survey you only get what you ask for

In depth information

No external validity

Results open to interpretation

Demand characteristics: People know they are being recorded/films. So people may not want to state their true feelings

Dominant personalities

Settings and procedures

Informed consent: full disclosure, individual comprehension, competence

Privacy: public/private settings

Confidentiality

Anonymity

Research Hypothesis: an empirically testable statement derived from theory. It is stated in terms of the population. It holds for the whole population, but we only have a sample. 

Null Hypothesis: the opposite of the research hypothesis given as a fixed value.

Assume the Null: to derive a sampling distribution. A mathematical probability distribution, we do not find it empirically

Determine the Critical Region: 5% or less

Calculate a test (sample) statistic

One and Two Tailed Tests: One tailed test: a statistical test where extreme results leading to rejection of the null hypothesis can be located at only one tail. Two Tailed Test: A statistical test where extreme results leading to rejection of null hypothesis will be located at both left and right tails

2 Error Types: Type I error is to rejecting a true null. Type II is failing to reject a false null.

Make a decision - infer about your research hypothesis.