Alongside theories, research methods shape academic disciplines such as communication. Whereas theories determine the subject matter (i.e., the part of reality a discipline is looking at), methods determine how a discipline gathers information about its subject matter. Which methods are acceptable and how methods are applied is subject to an ongoing debate and communication process within a scientific community. The correct application of research methods ensures that scientific results (1) are collected systematically; (2) are independent of the scholar who collected them; and (3) can be replicated by other scholars. In other words, research methods increase the credibility of results, and hence improve their quality.
This article provides an overview of the methods being used in communication research. It has links to nearly 60 more articles on research methods. This overview article will try to refer to most of them so that the reader can easily find more information on a given keyword. Besides the specific methods a vast body of literature provides interested readers with information on the theory of science and research focusing on the question of how we find evidence for our theories, whether we can eventually prove or falsify theories with the data collected, as well as how research depends on our conceptualization of science in general. Some of the most well-known authors in the field of general methodology are Karl Popper (1935) and Thomas Kuhn (1962). Popper’s basic argument is that theories in general cannot be verified, because we cannot know – neither in the present, nor in the past and future – about all events and facts relevant to the theory. Consequently, theories can only be falsified; good theories are those that survive plenty of falsification attempts. Kuhn argues that science does not develop in a linear fashion. Instead, scientific revolutions change the paradigm according to which research is conducted, thereby establishing a new view on the types of problems a discipline should address and the way of building theories about them.
Research methods can be grouped in three sub-fields: (1) data collection; (2) data analysis; and (3) study design. The kind of method through which a researcher collects data depends on whether he or she uses quantitative methodology or qualitative methodology. Quantitative approaches collect data from a large number of subjects, and are usually aimed at describing characteristics of, and trends in, the general population (e.g., a country) or at identifying differences between sub-groups (e.g., men vs women). Most of the time, only few attributes of subjects are measured. Typical studies with quantitative methodology are representative surveys among the general public. Qualitative approaches collect data from only a small number of subjects, and are usually aimed at describing individuals or small groups according to their behavior in an area relevant to communication, for example how politicians use the Internet. Most of the time, many attributes of subjects are collected in order to describe the complex nature of that subject’s behavior or attitudes. Studies with qualitative methodology are often in-depth interviews with individuals who are typical of the study’s subject.
Data collection can be conducted with (1) self-characterization of respondents; (2) observation of respondents’ behavior; (3) analysis of media or communication content; and (4) measurement of physiological parameters.
Respondents in interviews characterize themselves either in front of an interviewer who is asking questions or in written questionnaires (mail survey, online survey). The interviewer will ask questions either via the telephone or face to face. Telephone surveys have become extremely popular because they are fast, inexpensive, and easy to control (all interviewers are located in one spot). Computer-assisted telephone interviewing (CATI) makes it convenient to design, conduct, and analyze telephone interviews. Face-to-face interviews have the advantage of a higher response rate and better illustration. Visual materials can be used to illustrate scales, ranges, and so on. However, face-to-face interviews are more expensive, and the effects of the interviewer on the quality of the interviews pose a potential problem. Mail surveys usually suffer from a very low response rate, unless the researcher notifies respondents in advance and highlights the importance of the study. The term survey is used to describe mainly face-to-face interviews or telephone interviews with samples of the general population of a country. Professional survey-research institutes (e.g., Gallup or Roper in the US; Allensbach in Germany) offer representative results for those who are interested in public opinion.
Other than surveys, qualitative interviews are less standardized. The order and the wording of questions are to some degree dependent on the course of the interview. Instead of collecting a limited series of individual attributes (e.g., which party do you vote for, how many hours of television do you watch on a normal weekday), qualitative interviewers try to dig deeper and collect information about an individual as a whole. At the same time, there are neither a fixed number of people under investigation, nor a precise sampling procedure. Researchers often use theoretical saturation as a criterion to stop doing further interviewing. This means that interviews are conducted as long as the researcher has the impression that he/she still gets valuable information about the research topic.
Observation of subjects’ behavior is a multifaceted method. In general, researchers develop a coding scheme that counts the occurrence, frequency, and/or duration of certain behavioral instances. One of the most frequent observation methods is telemetry: devices built into the television set of a sample of television viewers enable researchers to identify the channel any subject is watching at a given point in time. As a result, ratings for a given television show can be computed. Other types of observations are conducted while people are engaged in group discussions. Their points of view, their nonverbal behavior, their arguments, and so on can be observed and counted.
Analysis of communication content can be conducted in several ways, depending on the type of study. Qualitative content analysis is used when data has already been collected from expert – ethnographic or historiographic – interviews. Quantitative content analysis is used to analyze larger amounts of mostly mass media messages like newspaper articles or television news. The results of content analyses may be used to make three types of inferences: (1) to the context in which a message was created; (2) to the motives and intentions of the communicator (journalists, public relations people); and (3) to possible effects of media messages. All three inferences are only valid if additional data, such as survey results, interviews with journalists, etc. is incorporated into the study. Physiological measurement (still) plays a minor role in communication research. Examples of fields in which this methodology is used are media use and reception as well as the effects of media violence. Recipients’ heart rates, electrodermal activity, or blood pressure are used to identify physiological arousal, which in turn can help to predict phenomena such as involvement with a media message, activation, and/or processing quality.
Whether data was collected using qualitative or quantitative methods, it needs interpretation and analysis. There is a whole spectrum of methods to do this. In the qualitative realm, the collected texts are reordered, interpreted, and reconstructed in order to find general ideas behind the text surface. The qualitative analysis is a combination of inductive and deductive processes. Theories and hypotheses are developed, compared against the textual material, rephrased, and compared again.
Quantitative data is mostly analyzed using a wide spectrum of statistical procedures and tests. These can be divided into descriptive and explanatory statistics. Descriptive statistics include the central tendencies of variables in a sample, such as arithmetic means, median, or standard deviation, as well as the distribution of frequencies for a given variable, such as the percentage of voters voting for each of a country’s parties. Explanatory statistics are designed to test the relationship between two or more variables and try to identify causal orders. They include group comparisons (t-test, analysis of variance) and predictions of a dependent variable by one or more independent variables. Most of these analyses are parametric – they are based on statistical distributions, mainly the so-called normal distribution. Less frequently used are different types of nonparametric analysis.
The adequate application of statistical procedures requires knowledge about the measurement level on which data was collected. Variables on the nominal level have a limited number of values that are mutually exclusive. A typical variable on the nominal level is gender. It has two values: “male” and “female.” Each observation in a data set has a value for gender, either one or the other. Nominal variables with two values are called dichotomous; when they have more than two values they are called polytomous. Variables on the ordinal level of measurement have a limited number of values that can be arranged in a rank order (the best, the second best, the third best, etc.). Grades in school are an example of rank-order data. Subjects in a sample can either have a unique rank each or can be tied, i.e., they share a rank. Variables on the interval level have values with equal distances. The interval level is most often found with Likert scales.
One of the most important issues in social sciences is the question of causality: in a complex social environment, how can we be sure that a given phenomenon A is the cause for another phenomenon B? Can it not be the other way round, i.e., B causes A, or even more likely that A and B are caused by a third phenomenon, C? The only study design that yields true causal relationships is the experimental design. Experiments are built upon two groups of subjects: the experimental group and the control group. These groups are created by dividing a set of subjects into two halfs, using either randomizing or matching procedures. As a result, the groups do not differ systematically in any variable. Ideally, all attributes of the subjects are distributed equally. Given this, the two groups receive a different “treatment.”
For example, the experimental group is presented a violent media stimulus, the control group a non-violent. The different treatments form the independent variable. After presentation, subjects’ reactions are measured, e.g., their aggressive tendencies. These reactions form the dependent variable. If the two groups differ in their reactions, the cause must be the different treatments, as we assume that the two groups are identical in any other regard. This can only be stated if there are no other differences between the two groups; statistically speaking, there are no error variables. Many error variables can occur during the presentation. Consequently, researchers try to standardize the exposure to the stimulus material as much as they can. This leads to high degree of internal validity. On the other hand, this procedure also creates rather artificial situations in which subjects are likely to behave differently from natural situations. Thus, the degree of external validity is low. The interpretation of the results of this experiment would claim that violent media content leads to aggressive tendencies among its viewers.
The experimental design can also be used outside the “laboratory.” Field experiments and natural experiments are conducted in the social environment of the subjects. It is more difficult to find two parallel groups of individuals, and it is even more difficult to control for error variables that occur during presentation. These types of experiments have a higher degree of external validity at the cost of lower internal validity. Any other non-experimental design does not allow for a causal interpretation of the relationship between two or more variables.
Communication processes occur over time. As a consequence, it is often not appropriate to investigate subjects’ communication behavior only once at a given point in time (cross-sectional analysis). A repetition of investigations over time is called longitudinal analysis. This kind of analysis can either be conducted as trend study or as panel study. Trend studies are often found during election campaigns. Surveys among samples of the electorate are conducted on a regular basis, using a “fresh” random sample every time. The results can be used to show developments in party or candidate preferences over time. However, trend analyses do not show whether individuals have changed their opinions and/or preferences. This can be shown by using panel studies. In these studies, the same subjects are surveyed at different points in time. Panel surveys, therefore, do not only show the gross development of public opinion, they also show how individuals have changed their views and what might have caused the changes. Problems arise regarding so-called “panel mortality,” i.e., respondents refusing to participate in the study any longer and dropping out of the panel. Increasing the time period of investigation increases panel mortality as well. In addition, panel studies are more expensive because panels need a lot more administration.
Another type of design is called Delphi Studies. In order to get assessment about future developments, researchers conduct mail or telephone surveys with experts in a certain field. Their responses are aggregated and sent out to the experts once more. Experts can then revoke their original responses in light of the results, i.e., the responses of the other experts. It is assumed that the results converge after the second survey, which in turn leads to more valid predictions.
Surveys and questionnaires often include standardized measures, mostly scales that explore a certain construct such as personality traits, attitudes, or person perceptions. Standardized measures are pre-tested and validated. Researchers therefore know about the distribution of the measures in a given population. A good overview of such measures is given by Rubin et al. (1994).
Quality Measures of Research
The quality of research methods is also an issue of methodological consideration. Systematization and objectivity have already been named as prerequisites of good research. They make given results independent of the particular researcher’s views or expectations, and enable other researchers to repeat the study, as long as all steps of a study are laid open in the publication. Particularly in the area of quantitative methodology, two further measures of quality are applicable: reliability and validity. Reliability indicates whether categories in a code-book (content analysis), questions in a survey, or categories in observation sheets measure a stable construct or not. The most common form of reliability is test-retest-reliability. After a certain interval, the same subjects are being measured again. If the first and the second measurement lead to the same results, the corresponding instrument (code-book, questionnaire) is reliable. Other forms of reliability include parallel-test or split-half-reliability. The degree of reliability varies between 0 (not reliable at all) and 1 (completely reliable). Without reliable measures, any conclusion or interpretation might be misleading or at least dependent on the situation in which the data was collected.
Validity refers to the question of whether a measured construct really measures what it intends to measure. In other words, the operationalization, i.e., the translation of theoretical concepts into measurement operations, should be valid. A good example for missing validity is the attempt to measure the construct “intelligence” by the perimeter of people’s heads. It is obvious that head perimeter has nothing to do with intelligence. The validity of a measure is very difficult to obtain and determine. Mostly, researchers rely on face validity using prior knowledge or common sense to estimate validity. Criterion validity uses an existing measure to validate a new one. Intelligence tests are often validated by using teachers’ judgments of their students’ intellectual capability. The question arises, then, of whether such judgments are valid themselves.
Another quality measure is the generalizability of results. As most research is conducted using random samples or nonrandom samples, results can only be generalized to a population if – and only if – a sample is representative of a certain population. Representativeness means that the distribution of attributes and variables in a sample is equivalent to the distribution in the given population.
Most of these quality criteria refer to quantitative methodology. It is much harder to establish such criteria for qualitative research. In particular, researchers have to prove that their results and interpretations are independent of their personality, or have to lay open how their own views and attitudes are related to their subjects. Triangulation is one way of increasing the quality of research. It includes employment of different methods, theories, or data sources on the same research topic in order to capture the phenomena in a broader way.
- Aneshensel, C. S. (2002). Theory-based data analysis for the social sciences. Thousand Oaks, CA: Pine Forge. Babbie, E. (1990). Survey research methods, 2nd edn. Belmont, CA: Wadsworth/Thomson Learning.
- Babbie, E. (2001). The practice of social research, 9th edn. Belmont, CA: Wadsworth/Thomson Learning.
- Denzin, N. K., & Lincoln, Y. S. (eds.) (2000). Handbook of qualitative research, 2nd edn. Thousand Oaks, CA: Sage.
- Flick, U. (2006). An introduction to qualitative research, 3rd edn. Thousand Oaks, CA: Sage.
- Frey, L. R., Botan, C. H., & Kreps, G. L. (2000). Investigating communication: An introduction to research methods, 2nd edn. Needham Heights, MA: Allyn and Bacon.
- Hinkle, D. E., Wiersma, W., & Jurs, S. G. (1994). Applied statistics for the behavioral sciences, 3rd edn. Boston: Houghton Mifflin.
- Katzer, J., Cook, K. H., & Crouch, W. W. (1998). Evaluating information: A guide for users of social science research, 4th edn. Boston: McGraw-Hill.
- Krippendorf, K. (2004). Content analysis: An introduction to its methodology, 2nd edn. Thousand Oaks, CA: Sage.
- Kuhn, T. S. (1962). The structure of scientific revolutions. Chicago: University of Chicago Press.
- Myers, A., & Hansen, C. (1997). Experimental psychology, 4th edn. Pacific Grove, CA: Brooks/Cole.
- Popper, K. R. (1935). Logik der Forschung: Zur Erkenntnistheorie der modernen Naturwissenschaft. Vienna: Julius Springer. English version pub. 1959 as The logic of scientific discovery. New York: Basic Books.
- Rubin, R. B., Palmgreen, P., & Sypher, H. E. (1994). Communication research measures: A sourcebook. New York: Guilford.
- Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-experimental designs for generalized causal inference. Boston: Houghton Mifflin.
- Tabachnick, B., & Fidell, L. (2007). Using multivariate statistics. Boston: Pearson Education.
- Watts, J. H., & Van den Berg, S. A. (1995). Research methods for communication science. Needham Heights, MA: Allyn and Bacon.
- Wimmer, R. D., & Dominick, J. R. (2003). Mass media research: An introduction, 7th edn. Belmont, CA: Wadsworth/Thomson Learning.