in some contexts, e.g., in wage rising discrimination, or in disparate application of
contractual terms, e.g., in house lending (Roscigno et al., 2009). (Bendick, 2007)
reviews more than 30 situation testing studies in employment discrimination in the
US, while (Rorive, 2009) covers the EU Member States context.
In correspondence testing , the data scarcity problem is mitigated by designing
paired ad-hoc fake resumes or application forms to be sent to advertised vacancies,
and by assigning to each of them a typical white American name or an African-
American sounding name (Arai et al., 2008; Banerjee et al., 2009; Bertrand &
Mullainathan, 2004; Carlsson & Rooth, 2007; Kaas & Manger, 2010; Neumark,
2010). Other grounds of discrimination have been covered with a correspondence
testing approach in job applications, including sex (Riach & Rich, 2006; Booth &
Leigh, 2010), obesity (Rooth, 2009), sexual orientation (Drydakis, 2009), ethnicity
(McGinnity et al., 2009).
Larger opportunities for data collection are offered by emerging Internet job ad-
vertisement services, known as e-recruiting (Booth et al., 2010; Edin & Lagerstrøm,
2006). The synthetic generation of resumes is tackled in (Lahey & Beasley, 2009)
by a parametric tool that mitigates the bias that is present in manually generated
CVs. The legal implications of possible discrimination in e-recruiting, as compared
to classical means of recruiting, are discussed in (Hogler et al., 1998). In addition,
contexts other than employment can be covered, such as discrimination in product
advertising in internet marketing (Doleac & Stein, 2010; Nunley et al., 2010), and in
on-line rental housing (Ahmed & Hammarstedt, 2008; Bosch et al., 2010; Friedman
et al., 2010; Hanson & Hawley, 2011; Taylor, 2010).
Field experiments construct control groups by matching similar persons and then
observing the outcome of a quasi-experiment in a natural environment, e.g., in a
job selection procedure. Empirical data from field experiments reflect a variety of
environmental factors: disentangling these factors may be difficult if not impossible.
Controlled experiments are conducted in an artificial environment, such as a labora-
tory, under tightly controlled conditions, including selection of treatment and control
groups and strict rules on their behavior and actions. On the one hand, the impact
of a specific factor can be evaluated by systematically varying it. On the other hand,
confounding variables and other extraneous stimuli can be minimized. Controlled
experiments are very useful to test the predictions of some theoretical model or to
pre-test the impact of some ruling or laws before their application. Also, controlled
experiments are repeatable, by definition, and less expensive than field experiments.
The main criticism against controlled experiments is that they suffer of lack of re-
alism, also called external validity . (Harrison & List, 2004) propose a taxonomy
of experiments. We refer to (Charness & Kuhn, 2011; Levitt & List, 2007) and
(R. M. Blank et al., 2004, Chapter 6) for an in-deep discussion on methodological
strengths and on the limits of generalizing results obtained from experiments.