Processing and Analysing Data

In one line

Collected data is not results: it has to be checked, coded, classified and tabulated before it means anything, and each of those four steps is a place where a study is quietly ruined.

In the wording a student can write in an exam: processing of data comprises editing, coding, classification and tabulation, by which raw responses are made complete, consistent and comparable and are arranged so that they can be counted; analysis is the examination of the tabulated data to establish magnitudes, distributions and relationships, using measures of central tendency and dispersion and, where appropriate, tests of association, in order to answer the research question and test the hypothesis.

Editing

What it is. Going through every completed instrument before anything is counted.

What to look for. Blanks; answers that contradict each other, such as a respondent who has never heard of legal aid but describes going to the office; answers outside the permitted range; illegible entries; and instruments completed in circumstances the investigator noted as unusual.

What may be done. Correct an obvious clerical slip; return to the respondent where possible; and mark an entry as missing.

What may never be done. Fill in an answer the researcher thinks the respondent would have given. That is fabrication, and it is invisible in the result.

Do it in the field, daily. An error found on day three can be repaired by revisiting; the same error found at the end cannot.

Coding

What it is. Assigning a symbol, usually a number, to each response so that it can be counted.

Closed questions code themselves, because the options were fixed.

Open questions have to be coded after collection, and this is the step that takes the time. Read a sample of the answers, build a set of categories from what is actually there, and then apply them to all the answers.

Three rules. The categories must be exhaustive, so everything can be placed. They must be mutually exclusive, so nothing can go in two. And there must be an other, which should be small; if other is a quarter of the responses, the categories are wrong and must be rebuilt.

Keep a codebook. A written record of every code and what it means, which is what allows another person to check the work and what allows the researcher to remember in month four what they decided in month one.

And code blind where you can. A researcher who knows which settlement an answer came from may code ambiguous answers in the direction they expect.

Classification

What it is. Grouping the coded data into classes for analysis: by attribute, such as sex or occupation, or by magnitude, such as age groups or ranges of expenditure.