Page 317 - CyberSecurityEssentialsEnglish

concept; that is a critical judgment. Mixing the levels makes conclusions stronger than the
text allows.

A data extraction form turns reading into structured data. It includes the identity of the
source, the question, the design, the sample, the setting, the intervention or exposure, the

outcomes, the measures, the limitations, the funding, and the location of the information.
The fields vary by project. Do not build a vast form collecting everything possible; restrict it

to what is needed to answer the question and build the synthesis.

When using AI with a full file, ask it to return the information with a page number, a section
heading, or a table reference, and to distinguish explicit text from inference. Then open the
location and verify. Page numbers may fail where a PDF is scanned or the file pagination

differs from the journal’s, so a short sentence or a table title can also be recorded. The aim
is that another person can find the evidence quickly.

A literature matrix places the studies in rows and the important fields in columns. It reveals
gaps and differences faster than separate files. Columns can be added for the method, the

context, the finding, the limitations, the quotation, and notes. But the matrix should not
replace the original texts; it is an organized index to them. Keep a copy of the source, its
bibliographic data, and its link or identifier.

Direct quotation must be kept separate from paraphrase. Use clear quotation marks and a

page number, and do not rely on the memory of a session or on a tool’s summary. When
paraphrasing, write the meaning from your own understanding and then compare it with

the source. Changing words while keeping the sentence structure is not scholarly
paraphrase, and it may lead to unintentional plagiarism.

In projects with dozens or hundreds of sources, a data dictionary helps. It defines each field,
its values, and its coding rules. For example: what counts as a university setting? How do we

record a mixed sample? How do we handle a study with more than one outcome? This
dictionary prevents decisions from varying between researchers and makes any automation
more consistent.

A tool can propose an initial coding of articles, or extract country names, sample sizes, or

types of instrument. But a sample must be tested manually and the errors counted,
especially in complex tables or with similar terminology. It is not enough for the cells to look
full; what matters is that they are correct and taken from the right place.

When two papers conflict, do not rush to choose one. Record the differences: did the

population change? Is the operational definition different? Is one longitudinal and the other
cross-sectional? Have the time, the technology, or the regulatory context changed? Much of

316