Theories and models form the basis for empirical inquiries. In empirical research we are not only interested in variations in our data (what we observe happening in the world), but we also want to test whether the data fits to our model or a theory. We also want to develop theories or even build new ones based on the previous ones and/or new empirical evidence.
Social scientists usually ask questions concerning humans’ social behavior, attitudes, and beliefs. How do people behave in certain social circumstances? Are there any regular patterns of behavior that can be repeatedly observed? For example, what attitudes are associated with a high degree of online participation? What makes people vote for certain candidates in presidential elections? What is the effect of media coverage? How do socioeconomic factors predict and influence media consumption? And so on.
In general, there are two strategies to follow when conducting quantitative research. The first, in which researchers formulate hypotheses on the basis of the previous research and test them against empirical data, is called confirmatory research. Confirmatory research confirms (or rejects) hypotheses. The other, possibly supplementary, strategy is exploratory research. Exploratory research is by definition exploration, a kind of adventure into the data. This method starts with the data and exploring in order to formulate hypotheses and theories based on the understanding derived from the data.
Some data analyzers may suggest that exploratory data analysis could be followed by confirmatory analysis: first the researcher explores the data, finds some interesting associations, then finds theory to support the observed variations, and finally performs a statistical test to verify what has just been found. However, there is a danger in this reasoning. The data can include correlations which do not really exist but are in the data just by chance. In theory, one can find arbitrary evidence from the data and develop a theory around it to “confirm” the observations. This is how human brains mostly work: we see patterns (“evidences”) around us and try to explain them. However, patterns are also seen where there is actually nothing at all going on!
This confirmation bias can be harmful in scientific research, and statistical methods are exactly the way of avoiding succumbing to it. Therefore, good scientific research is based on some type of theoretical reasoning, either taken as given and tested against empirical data or developed over the course of data exploration and carefully linked with an existing body of empirical research and theories.