Once the categorization and coding have been carried out, we move on to the analysis of the data themselves, depending on the specific technique of content analysis itself. In general, we proceed to establish the counting rules, once we have marked the presence of these categories in the units that make up our corpus: presence/absence, frequency, intensity, direction, order or contingency are values that can be operationalized.

It is advisable, to begin with, to carry out a pilot test. Pilot tests are used in surveys, experiments and content analysis. Its usefulness is to test the selection of the sample, the units of analysis and the design of the analysis sheet and thus to improve them before their definitive application. It is performed on a small selection of contents that are part of the sample. Any deficiencies that may appear will be recorded as incidences and thus the exclusion and sensitivity are refined. 

Likewise, some mechanism will be established to ensure the validity of the data. We can distinguish two types of validity:

1) Intersubjective validity= The same data is considered valid by two or more people when they apply the same instruments to validate it (quality of objectivity). Thus the researchers have the guarantee that the data collected by the whole team will be valid.

2) Intrasubjective validity= It fails when the same researcher modifies the criteria during the course of data collection. The statistical procedures Scott’s Pi index or Cohen’s Kappa (see Wimmer and Dominick, 1996) make it possible to calculate the level of differences in the data collected, to assess whether it is necessary to adjust the codebook and start again (if the variations are significant).

In general, and for the sake of simplicity, rather than formulas or coefficients, we prefer to speak of validity criteria.

1) Stability: The results must be stable and homogeneous over time, both when the same methods are applied in a longitudinal study and when a pre-test is performed and then the technique is applied to the entire corpus.

2) Reproducibility: The analysis is correct when it can be recreated in different situations (test-retest), and also when it is applied by several coders (test-test).

3) Accuracy: The analysis must conform to a previously established protocol, which all coders must respect, and is related to a norm or an agreement that must be produced both within the research group itself and according to the standards of the international scientific community. For example, taking similar analyses as a model.

Of these, specialists insist on the second, which is also called replicability. Klaus Krippendorf introduces the concept as a further characteristic of this procedure. Content analysis works with corpora that are understood to be "immovable", "insensitive to variation" (analysis of a literary text, for example). For this reason, in a selected and coded sample, applying the same procedures, the results must be the same, even if the researchers change.

Once this pilot test has been passed, we proceed to the real analysis. To this end, a protocol is established, which includes precise rules such as the description of the characteristics of the corpus and corpora, if any, of the units of analysis, categories and codes, their application, the subsequent treatment of the data and even the design of a database that allows us to systematize the analysis, always remembering that we are looking for patterns and variations.

Context, on the other hand, is very important in content analysis. According to Krippendorf, it contributes to conceptualize the portion of reality that originates the analyzed text. In the analysis of mass communication, the context can be the characteristics of the communicator, the effects of the message, the degree of public tension, the socio-political climate, the processes of value mediation. The context (surrounding conditions, antecedents, consequents, etc.) must be constructed by the analyst. The analyst’s task also consists in establishing the logical limits to the context. The purpose or objective of the inferences must be clearly stated. Content analysis provides vicarious information and formulates inferences. The analytic construct is provided by the relationship between the data and its context. The tests needed to validate the result are specified before starting the research. The methodology advances only if the results are validated. There is no place for singular or insufficiently conceptualized cases.

All the aforementioned procedure is summarized in this table:

One of the most attractive and productive techniques possibly applied to the study of the media is the proposal of the American professor David L. Altheide, which he calls Qualitative Document Analysis (QDA) or Ethnographic Content Analysis. Altheide, which he calls Qualitative Document Analysis (QDA) or Ethnographic Content Analysis, and which he has productively applied to media such as television or the Internet (see Altheide, 1987, 2000 and 2004). This approach aims to carry out an exhaustive monitoring of certain media content (discourse tracking) with the objective of transforming research questions into categories, which relates it to grounded theory. Altheide's proposal is both an interpretative and thematic textual analysis, which adapts to and takes into account media logic. It is based on both symbolic interactionism and structuralism. In symbolic interactionism because it understands that the ultimate meaning of any text or discourse lies in the interpretation of the cultural material by each actor (producers or users, for example). In structuralism, because it necessarily takes into account the cultural context of the messages.

The researcher himself from the University of Arizona, USA, defines his proposal as follows: "Ethnographic Content Analysis consists of reflexive movement between concept, development, sampling, data collection, data coding, data analysis, and interpretation". In reality, these are the same operations followed by any type of content analysis, quantitative or qualitative. The analysis itself, in Altheide's proposal as in many other types of content analysis, is carried out:

1) Comparing items.

2) Enumerating patterns, variations and trends.

3) Examining denotative and connotative changes.

4) Combining words and searches to find patterns and thematic recurrences.

Discourse tracking can be carried out using databases, Boolean searches and the possibility of adding or modifying categories.Altheide's qualitative or ethnographic content analysis, which aims to analyze the frequency and variety of messages and their characteristic features, or categories, is in turn based on: