Sometimes it may feel tricky to find out what analysis method to choose for one’s study. Indeed, all the components influencing on the selection, such as research questions, types of data, measurement levels, observed distributions, etc., are entangled. However, there is certain logic behind the selection criteria.
It’s good to start with research questions: what do you want to know and can your data answer on that question? How variables and observed values could inform you about the questions asked?
The table below includes some of the most common analytical settings and their related procedures. The first colum asks what type of variables do you have, what is their measurement level? The second is about the association that is studies, whether it is a linear correlation, between-groups mean difference, or difference in joint frequencies. The third column lists some of the statistics that can be used and the fourth a suitable statistical test for testing whether the observed results are not likely to be by chance alone. The last column suggests a type of visualization for the results.
Type of association
Two continuous variables x and y
Linear correlation between x and y
Mean, median, standard deviation
Correlation coefficient & coefficient of determination
The design of analysis most importantly consists of a determination of what are the sampling units and what are the units of analysis.
In sampling, researchers need to consider what the sampling unit (or an observational unit) will be. The researcher usually determines the population and takes a sample of it. For example, a population can be all the magazine covers of Time. As Time has been published weekly since 1923, there are about 33 x 12 x 4 = 4320 different magazine covers. This seems to be a large population, and therefore a sample can be taken.
How to determine a sample? The selected strategy and the sufficient amount of observations depend on the questions that are asked. One way would be to take a random sample from each decade and get a representative picture of the 20’s, 30’s, 40’s, etc. magazine covers. There are other options too, such as taking a sample from each editorial period.
Units of analysis
Let’s study the degree of violence in foreign news of different types of media. In this case, the sampling unit is a news story, although the inferences are made at the level of news media. What is the unit of analysis, i.e. what is observed?
The assessments of the degree of violence in stories are done by using three classes: 1 = the news story does not include any violence, 2= there is some violence, 3 = the violence is an essential part of the content. This is the unit of analysis, i.e. the violence in news representations. The unit of analysis can sometimes be rather a subjective as here, and not manifest in the classical sense (such as number of different genders appearing in pictures, etc). More objective measures could be constructed by counting some specific determinants of violence in news stories (cf. movie rating systems as an example of a coding scheme in this context).
From the picture below we can see that tabloids tend to have more violence in their foreign news coverage (almost 70% of all the contents on foreign news include some violence) than other news media. For example, in other news papers, it is less than 20%. On the other hand, violence in online news tend to take a more central role than in traditional versions, as well as in television and news papers.