What is reliability ? Explain the different tests available to social science researcher to establish reliability.

What reliability is

• In survey and test work, reliability is the share of variance that is not random error.
• Emile Durkheim already needed stable classification of deaths before suicide rates could be compared. Unstable coding is an reliability failure in official data.
• Qualitative work uses a cousin idea: inter-coder agreement and stable use of a codebook, not a claim that meanings never shift.

Test-retest

• The same test is given twice to the same persons after an interval.
• Use when the trait should be stable, such as a literacy item, not a mood item after a riot.
• Memory and real change both threaten the correlation.

Parallel or alternate forms

• Two versions of the scale are built to the same blueprint and both are given.
• Use in examinations and long batteries to reduce item memory.
• Costly to write well.

Split-half and internal consistency

• The test is split, scores are correlated, and Spearman-Brown may correct the half-length.
• Cronbach’s alpha estimates consistency among items of a Likert battery.
• Use for attitude scales on, say, gender-role items in a student sample. Alpha is not a test of unidimensional truth, only of hanging together.

Inter-rater or inter-coder reliability

• Two observers code the same focus-group transcript or Census occupation string.
• Cohen’s kappa or percent agreement is reported.
• Essential for content analysis, courtroom files, and open-ended NFHS-type comments.

Conditions and cautions

• High reliability does not prove that “caste” or “work” was well conceptualised. That is validity, and Weber’s meaning problem.
• Official series such as the Census need stable definitions across decades; a definition change looks like social change if reliability of the series is ignored.
• Hidden populations reached by snowballing cannot borrow NFHS reliability claims without a design argument.