Mass Communication Research
Hypothesis Testing
INDEX SYLLABUS SCHEDULE e-MEDIA COMM-STOP

I. Research Questions or Hypotheses?

Research begins with tentative generalizations regarding the relationships between two or more variables. As stated before, a hypothesis is a theoretical statement of relationship between two or more variables that can be directly tested. Research questions are the more generalized statements about a problem for study.

For instance, a researcher (you) may want to know what group of people in society spends their time searching the Internet for their daily news. So a research question may be phrased as:

What factors influence how much time a person spends surfing the Internet for news?

It would be necessary to gather demographic information to answer the question. Demographic factors, or variables, such as age, gender, education levels, income levels, computer access, etc. would provide a snapshot of those people who surf the Internet for news. A survey would probably ask respondents close-ended questions such as the following:
  1. Approximately how much time do you spend surfing the Internet?
  2. Approximately how much of that time is spent searching for news?
  3. What sites do you use for news?
Each of these questions could be paired with responses that can be measured from the nominal to the ratio level. Once the information is gathered, hypotheses can be used to examine if any relationships are worth reporting. In our example, some hypotheses may include:
  1. Males spend more time surfing the Internet than females.
  2. People between the ages of 18 and 25 spend more time surfing the Internet for news than people older than 25.
  3. People who make more than $50,000 annually spend less time surfing the Internet for news than those who make less.
  4. College graduates spend more time surfing the Internet for news than non-college graduates.
II. Purposes of hypotheses
  1. They provide direction for the study.
  2. Eliminate trial-and-error research.
  3. Help rule out intervening and confounding variables.
  4. Allow for quantification of variables.
It is important to use operational definitions. If a researcher is developing hypotheses about violence and television exposure, the researcher needs to have violence and television exposure defined for the study.

In our example, we would need to operationally define "searching" for news. Does it only include those times a person uses a "search" engine? Is it going to the same news source on a daily basis? What qualifies as a news source? Is the "Drudge Report" the same thing as The New York Times?

III. Criteria for good hypotheses
  1. Compatible with current knowledge.
  2. Logically consistent.
  3. Succinct.
  4. Testable.
IV. The null hypothesis

The null hypothesis asserts that the statistical differences or relationships under analysis are due to chance or random error. The null hypothesis is the logical alternative to the research hypothesis. For example:
Research hypothesis: 
People who own a computer spend more time searching the Internet for news than people who don't.

Null hypothesis: 
Computer ownership does not determine the amount of time spent searching the Internet for news.
V. Which inferential statistic?

How will you know which statistic you should choose when confronted with a hypothesis that you have made? When testing hypotheses, it is necessary to look at the dependent variable's level of measurement to determine the correct inferential statistic that will be used. In the communication research class, we consider the Chi-Square, t-Tests, ANOVA and Correlation statistics.

VI. When do you accept or reject?

Researchers usually talk in terms of accepting or rejecting the null. Even though the researcher states a "research" hypothesis, he or she will either accept or reject the "null" hypothesis when reporting results. When the null is rejected, the research hypothesis would be accepted.
Region of Rejection - the proportion of the area in a sampling distribution in which the null hypothesis would be rejected. It is defined by the level of significance the researcher chooses.
This is where SPSS and testing for statistical significance come into play. In communications research, the significance level to determine findings worth reporting is usually set at 0.05. This means that the difference (or relationship) seen in the data would occur in similarly-sized randomly-drawn samples where there was no difference less than 5 percent of the time. Because the chance of finding such a sample is low, we conclude that there is a difference in the population.

We have tested the null hypothesis — that the differences (or relationships) can be attributed to chance — and a .05 significance level means that we can reject the null hypothesis. We think the findings are attributable to something other than chance, but the test doesn't tell us what it is. In fact, we fairly often see findings at the .05 level that are not what we think they are. This is where the notion of internal validity comes in. If the relationships are not what they seem to be, then our findings lack internal validity.

It's important to know if you're testing a one-tailed or two-tailed hypothesis. If it's a one-tailed hypothesis, you will divide the significance level given in most SPSS output (you get to choose when running correlation). Remember:
  1. A one-tail hypothesis predicts the results will only fall in one direction. It is usually stated with adjectives including "more," "less," "higher," "lower," etc.
  2. A two-tail hypothesis does not predict the results one way or another. It is usually stated as "There are differences ..."
VII. Interpretation

When interpreting Chi-Square, you should discuss differences in percentages. When interpreting t-Tests or the ANOVA, you should discuss differences in means. When using Correlation, you should interpret the magnitude of the correlation coefficient to discuss relationships. (You can press any inferential statistic listed to go to a more detailed explanation).

VIII. Error

Though we've correctly tested our hypothesis, it does not guarantee that our decisions will be right. A researcher must be wary of error.
  • Type I Error: the rejection of null that should be accepted.
  • Type II Error: the acceptance of the null that should be rejected.
Researchers should state that a hypothesis is "supported" rather than "proven." It's impossible to "prove" something exists, though our data support our conclusions. We may get different results with different data. For more information on hypothesis testing, read Dr. Miller's essays on the Logic of Statistical Hypothesis Testing, and the Steps in Statistical Hypothesis Testing.

If you don't understand something in this Web note, please e-mail Dr. Sitton.

INDEX SYLLABUS SCHEDULE e-MEDIA COMM-STOP
©M. Mark Miller & Ronald W. Sitton 2009
Revised 092811 — http://www.uamont.edu/FacultyWeb/sitton/crz/mrea/hyptst.html