9.2 Field experiments Not all experiments are done in laboratory-type situations. One form of experiment is called a field experiment. An example is a study by Piliavin, Rodin and Piliavin (1969) of 'bystander apathy' in the New York subway system. They set up a situation in which a presumed ordinary passenger got on to a train and then, as the train was going along, pretended to collapse on to the floor of the carriage. The experimenters then recorded whether the person was given help, by whom and how long it took for help to be offered. Contrary to expectations that the New York subway is one of the worst places in the world for bystander apathy, the experimenters found, in this case, that help was offered, usually within two minutes. The fact that they are involved in an experiment is not revealed to the subjects under study.
This form of experiment, however, raises the fundamental issue of control (see Section 9.1.4). There is no way in which the field experimenter can begin to allow for all the possible control variables and this makes any causal analysis fraught with difficulty.
Arnfinn Midtbøen (2015) conducted a field experiment of ethnic discrimination in the Norwegian labour market. Fictitious application letters and résumés were sent to 900 job vacancies in the Oslo area. Each of the 900 posts received a matched pair of male (476 posts) or female (424 posts) applicants that differed in essence only by the name of the respondents: a traditional Norwegian name matched by a Pakistani name.
The applications were similar in age (both were 25 years old) educational attainment, work experience and computer skills. The résumés and cover letters were written in fluent Norwegian, and both candidates reported schooling from Norwegian educational institutions and former work experience in Norwegian firms. The field experiment thus measured employment discrimination of children of Pakistani immigrants. The résumés and cover letters, based on pre-made templates and slightly adjusted to fit each job posting aesthetically, such as different fonts and order of the qualifications listed, but these were assigned randomly. Thus, Midtbøen (2015, p. 199) claims that:
Consequently, any difference in call-back rates between the applicants is attributed to their different names and interpreted as an effect of ethnic discrimination.
The results confirmed that ethnic minorities are discriminated against as 42.8% of the 'Norwegian' applicant were offered an interview compared 31.9 % of the 'Pakistani' applicants: the difference of 10.9% is statistically significant (p < .01) (see Section 220.127.116.11).
The discrimination differs across sectors: there are large and statistically significant differences in applications for jobs as drivers, warehouse workers and accounting assistants but small and non-significant differences for pre-school teachers, nurses and public consultants.
The results confirm 40 years of similar research in twelve countries (Pager 2007; Riach and Rich 2002). What it does not explain is why the differences occur. Midtbøen thus followed up the field experiment with in-depth interviews with 42 employers. The interview data suggested:
the number of applicants influences the degree to which employers can 'afford' to discriminate, indicating that a small pool of relevant candidates provides less room for discrimination. Second, there are large differences in what formal requirements mean in practice, even between occupations requiring several years of college education. The qualitative 'tracing' of employment processes indicates that while formal requirements in some occupational contexts are predefined and absolute, other contexts leave the question of what constitutes a qualified candidate open to employers' discretion. In terms of hiring outcomes, discretionary decisions appear to increase the probability for discriminatory behaviour. Third, formalized recruitment procedures seem to decrease the salience of ethnicity. The bureaucratization of procedures minimizes individual differences, providing less room for decision-making based on biased first impressions. (Midtbøen, 2015, pp. 208–9)