A census involves (attempted) complete listing/enumeration, collection and analysis of data from every unit, case, or member of a group/population
(Saunders, et al., 2018). It’s characterised by “universality, simultaneity of information, and individual enumeration” (Nirel & Glickman, 2009, p. 539). An
example is the national population census, which are undertaken by many countries often after 10-year intervals (Nirel & Glickman, 2009). A census is
expensive, time-intensive, and unavailable for some research questions/objectives (Saunders, et al., 2018). A sample is a subset of the population. It makes a
valid alternative when a census is impractical and/or infeasible. Samples may, however, be biased or unrepresentative of the population and can be
impossible in some cases (Bryman, 2015).
Non-probability sampling offers convenience, practicality, and potential to collect data rom the most insightful respondents (Daniel, 2011; Bryman, 2015).
According to (Bryman (2015), however, the subjectivity involved, however, introduce sampling bias and systematic errors, as well as undermine the study’s
external validity. Probability sampling offers less bias (Saunders, et al., 2018). Further, by estimating confidence levels, it’s possible to estimate the accuracy
with which the sample reflects population parameters, effectively making it possible to generalise findings to the population. Non-probability sampling may
be infeasible/impractical (Saunders, et al., 2018).
Secondary data refers to existing data, collected for other purposes (e.g., domestic violence abuse data from Office for National Statistics (2022), but which
may be analyse to generate more or different insights. Primary data is newly-collected, for the purposes of satisfying specific questions/objectives
(Saunders, et al., 2018). According to Bryman (2015), secondary data is economical, convenient to gather, and allows for triangulation of sources. It may,
however, have to be extracted from diverse datafiles, contain errors, manipulations/inaccuracies, and gaps rendering it incapable of answering questions at
hand (Saunders, et al., 2018).
How do you reconcile probability sampling and generalisability findings with non-positivist philosophical beliefs that do not consider generalisability as
desirable?