Page 327 - CyberSecurityEssentialsEnglish

the context, and the assumptions. Do not carry over the model’s interpretation before
reviewing it against the logic of the method.

Choosing between a test and a model is not a matter of phrasing. It depends on the type of
variable, the study design, independence, distribution, clustering within people or

institutions, and whether the aim is descriptive, predictive, or causal. A tool can prepare a
decision tree of questions, but the answers come from understanding the data and the

method. And where the expertise is not available, a statistician should be involved rather
than hiding the uncertainty inside generated code.

In qualitative research, a tool can help format texts, detect initial names or themes, propose
disconfirming examples, and compare code definitions. But it does not live the participants’

context and does not always understand irony, power, or cultural meaning. Building the
question, the relationship with participants, reflexivity, and interpretation must remain the
researcher’s responsibility, consistent with the literature on thematic analysis and

qualitative coding [26][27].

A hybrid coding route can be followed: the researcher reads a sample and builds a draft
codebook, then the tool tests the application of the definitions on selected texts, the
differences are reviewed, the rules are adjusted, and the route is applied to a larger portion

with periodic checking. Numerical agreement alone should not be used as evidence of
quality; everyone may agree on a shallow and weak coding.

When analyzing interviews or sensitive data, privacy is fundamental. Removing a name is not

enough if the text contains a rare occupation, a location, or an event from which the person
could be identified. Ethical approval, the data management plan, and the contracts must be
respected. In some cases an external tool may not be used at all, and local processing or

human analysis is the appropriate option.

In research translation, a tool may help translate interviews or questionnaires, but
translation sometimes changes meaning. It is better to keep the original text, document who
translated it, review sensitive terminology, and perhaps use back-translation or bilingual

reviewers when language is part of the phenomenon under study.

Analysis must be separated from presentation. A tool may propose an attractive chart that
truncates an axis, hides variation, or uses misleading colors. Ask: does the figure answer a
question? Does it show the sample size and the uncertainty? Does it use the correct unit?

And can a reader interpret it without the text? Visual productivity does not justify distorting
the data.

326