- H16: Most practitioners agree that requirement coverage is more important than code coverage (average Likert score = 4.00)
- H17: Most respondents agree or strongly agree that test cases should be well modularized (average Likert score = 4.62)
- H18: More than 96% of the respondents agree that test cases should be readable and understandable
- H19: 57.42% of the respondents agree or strongly agree that test code should be simpler than the tested code, while 16.41% indicate their disagreement or strong disagreement (average Likert score = 4.20)
- H20: Most practitioners agree or strongly agree that test code should be designed with maintainability in mind (average Likert score = 4.16)
- H21: More than 190 practitioners agree or strongly agree that traceability links should be maintained between test cases, code and requirements (average Likert score = 3.97)
- H22: A total of 215 respondents agree or strongly agree that a test case should attempt to break a functionality (average Likert score = 4.11)
- H23: The majority of respondents agree that testing even the simplest things is valuable (average Likert score = 3.89)
- H24: Add new testcases for fixed bugs (average Likert score = 4.40)
- H25: Most of our respondents agree that assertions are a crucial part of test code and can be helpful in detecting subtle errors (average Likert score = 4.51)
- H26: A large number of our respondents (i.e., 200) agree or strongly agree that commenting test code with common errors and possible causes is a good idea (average Likert score = 3.98)
- H27: Most practitioners we survey agree that a good test case should be deterministic and produce the same output every time it is run (average Likert score = 4.07)
- H28: Almost all our respondents agree that test cases should be side effect free
- H29: Several common testing frameworks like JUnit provide the functionality of adding tags to test cases. Most of our respondents agree or strongly agree that the use of such tags to indicate, for example, fast or slow tests, is helpful (average Likert score = 3.93)

Полный текст статьи - "Practitioners’ Views on Good Software Testing Practices"

Меня немного удивил результат для гипотезы H24. Я знаю о такой практике, но не думал, что так много людей считают её правильной. Если так получилось, что тесты пропустили баг, то лучше сделать постмортем, попытаться понять что не так в ваших тестах, процессах или требованиях и что нужно сделать чтобы этого не повторилось в будущем. Систематическое создание теста для исправленного бага будет делать ваше тестирование бессистемным.