There are two discrete steps in performing a cultivation analysis. First, descriptions of the media world are obtained from periodic content analyses of large blocks of media content.
The result of this content analysis is the identification of the messages of the television world. These messages represent consistent patterns in the portrayal of specific issues, policies, and topics that are often at odds with their occurrence in real life. The identification of the consistent portrayals is followed by the construction of a set of questions designed to detect a cultivation effect. Each question poses two or more alternatives. One alternative is more consistent with the world as seen on television, while another is more in line with the real world. For example, according to the content analyses performed by Gerbner and colleagues (1977), strangers commit about 60% of television homicides. In real life, according to government statistics, only 16% of homicides occur between strangers. The question based on this discrepancy was, “Does fatal violence occur between strangers or between relatives and acquaintances?” The response “strangers” was considered the television answer. Another question was, “What percentage of all males who have jobs work in law enforcement and crime detection? Is it 1% or 5%?” According to census data, 1% of men in real life have such jobs, compared with 12% in television programs. Thus, 5% is the television answer.
Condry (1989) points out that the cultivation impact seems to depend upon whether respondents are making judgments about society or about themselves. Societal-level judgments, such as the examples just given, seem to be more influenced by the cultivation effect, but personal judgments (such as “What is the likelihood that you will be involved in a violent crime?”) seem to be harder to influence. In a related study, Sparks and Ogles (1990) demonstrated a cultivation effect when respondents were asked about their fear of crime but not when they were asked to give their personal rating of their chances of being victimized. Measures of these two concepts were not related. Related findings were reported by Shanahan, Morgan, and Stenbjerre (1997), who found that TV viewing was associated with a general state of fear about the state of the environment but not related to viewers’ perceptions of specific sources of environmental threats.
The second step involves surveying audiences about their television exposure, dividing the sample into heavy and light viewers (4 hours of viewing a day is usually the dividing line), and comparing their answers to the questions that differentiate the television world from the real world. In addition, data are often collected on possible control variables such as gender, age, and socioeconomic status. The basic statistical procedure consists of correlational analysis between the amount of television viewing and the scores on an index reflecting the number of television answers to the comparison questions. Also, partial correlation is used to remove the effects of the control variables. Alternatively, sometimes the cultivation differential (CD) is reported. The CD is the percentage of heavy viewers minus the percentage of light viewers who gave the television answers. For example, if 73% of the heavy viewers gave the television answer to the question about violence being committed between strangers or acquaintances compared to 62% of the light viewers, the CD would be 11%. Laboratory experiments use the same general approach, but they usually manipulate the subjects’ experience with the television world by showing an experimental group one or more preselected programs.
Measurement decisions can have a significant impact on cultivation findings. Potter and Chang (1990) gauged TV viewing using five different techniques: (1) total exposure (the traditional way used in cultivation analysis); (2) exposure to different types of television programs; (3) exposure to program types while controlling for total exposure; (4) measure of the proportion of each program type viewed, obtained by dividing the time spent per type of program by the total time spent viewing; and (5) a weighted proportion calculated by multiplying hours viewed per week by the proportional measure mentioned in the fourth technique.
The results showed that total viewing time was not a strong predictor of cultivation scores. The proportional measure proved to be the best indicator of cultivation. This suggests that a person who watches 20 hours of TV per week, with all of the hours being crime shows, will score higher on cultivation measures of fear of crime than a person who watches 80 hours of TV a week with 20 of them consisting of crime shows. The data also showed that all of the alternative measures were better than a simple measure of total TV viewing.
Potter (1991a) demonstrated that deciding where to put the dividing point between heavy viewers and light viewers is a critical choice that can influence the results of a cultivation analysis. He showed that the cultivation effect may not be linear, as typically assumed. This finding may explain why cultivation effects in general are small in magnitude; simply dividing viewers into heavy and light categories cancels many differences among subgroups. Diefenbach and West (2001) offer another insight into possible ways of measuring the cultivation effect. In their study of the cultivation effect, they found no relationship between TV viewing and estimates of murder and burglary rates in society when using the traditional regression model. However, when they used a different form of regression analysis, one based on non-normally distributed dependent variables, they detected a cultivation effect.
More recent methodological investigations include those of Hetsroni and Tukachinski (2007) who found that classifying viewers based on both their estimates of the occurrences television and real-world phenomenon provided clearer depictions of a cultivation effect and Van den Bulck (2003) who examined if the mainstreaming impact could be explained by regression toward the mean.