Verification

Verification checklist

The verifier (someone else in the team who did not enter the questionnaire) follows a checklist to make sure all the content has been added and is in line with our protocols.

Summary tables

As an additional check, summary tables can be used. Each construct on the build page (e.g. Conditions) provides a summary of that construct. By searching and highlighting the prefix used (e.g. c_ in conditions), it will indicate which labels have the correct prefix.

GitLab

To more quickly and accurately complete bulk checks, particularly for large questionnaires with hundreds of constructs, the archivist_verify GitLab pipeline can be used. This checks for issues such as labelling (e.g. all question items should start with qi_) and empty constructs (e.g. conditions should have at least one construct inside them). A full list of checks can be found in the readme.md on GitLab. The pipeline outputs .csv files highlighting the potential issues. As the pipeline does not have the capability to take into account the many complex rules and protocols we have at CLOSER, it is important the verifier checks that the errors are valid before adding them to the verification notes.

For guidance on how to set up and run the archivist_verify pipeline, see here

Listening

This method is no longer used, as we looked into the number of hours spent listening and the number of additional mistakes the verifier spotted as part of this process. The length of time the verifier spent listening to the questionnaire did not justify the number of additional mistakes spotted, so this method was discontinued. The old process is highlighted below.

As part of the verification process for paper questionnaires, the entered questionnaire is listened to. The advantage of this over proof reading is that the software cannot be fooled by word shape and letter recognition. An example picked up early on was that someone had entered martial instead of marital. This would be a difficult error to catch by reading as the reader tends to use context of what they are reading. It is entirely possible to read 'Did you have any martial problems?' in a section on marriage and miss the error because the structure of the words is so similar. The software, however, clearly reads what is there and this is evident to the verifier. Additionally, if the error has produced a word that is not readable, the software reads it out letter by letter, bringing it to the attention of the listener. Given that the entry process is to replicate what is on the page, the risks of common grammatical errors such as two, too, to is low. The entry does not involve making a judgement, just reproducing what is on the page. If the usage is wrong on the original, that would be replicated.

Text to speech software: https://ttsreader.com/. Copy and paste the entered questionnaire from Archivist into the text box and press the play button.