Citizen profiling during weapon controls in Amsterdam: Observing potential biases

Published on 31/01/2023

by Hans Myhre Sunde & Mara van Dalen
Netherlands Institute for the Study of Crime & Law Enforcement (NSCR)

This blog is partly based on an already published blogpost.


In September 2022, the Mayor’s office of Amsterdam decided to re-introduce a politically heated and previously abandoned police practice – the random weapon controls. In the past, such weapon controls were criticized for being biased towards minorities and/or areas with predominantly minority populations. Thus, it was not without tension our research group set out to observe this practice in the real-world. To avoid issues of reactivity, we found it necessary to conduct covert observations to not influence the officers on site. The American Psychological Association states that such observations, when made in public, is ethically justified when it is not harmful to the observed. One of the ways to ensure this, is to anonymize the observed individuals by employing strictly quantitative coding instruments. In our case, we only recorded static characteristics based on our visual assessment of age, gender, and ethnicity. Our ethnicity variable was derived from a profiling-study conducted in France by Jobard and Levy (2011), and included the categories White, Arab, Black, Indo-Pakistani and Asian.

Then, we asked ourselves – is such research even possible? Can we really “spy” on the police and remain unnoticed? So, we went out to find out – and agreed in advance that the observations should stop if/when any officer spotted us. We attempted to stay “hidden” by performing everyday activities such as drinking coffee, eating pastries or talking on the phone. Doing nothing would put you “out of place” and the police would be more likely to notice. Based on this, we concluded that we could, reasonably well, instruct the research team on how to stay unnoticed.

In total, we observed six days of controls before the police unfortunately closed down the pilot due to “unauthorized practices”. Practically speaking, we were continuously informed via WhatsApp by the police coordinator what areas they would conduct their controls in. Researchers from the team made their way to the location, and initially observed the area to get an overview of what went on. Quickly after, we found spots in which it would be possible to observe the on-goings without the police noticing. Once in place, the observers coded people as they passed by into the search area – roughly every 3rd person – and recorded their age, gender, ethnicity, if they were with a family (this variable was included because families should not be searched), and whether or not they were searched.

One of the main strengths of observing the real-world is that it allows us to use a real-world control group instead of drawing on population or neighborhood data. We can therefore relatively accurately measure how often individuals from each of the 5 ethnic categories were searched, and then control for their relative proportion of the population on-site. For unbiased searches, then, the proportion of individuals from each ethnic group searched, would match the proportion of individuals from each ethnic group present in the area at the specific time.

Our observations lead to two main lines of findings that we described in the research report (Lindegaard et al., 2023). First, we observed three different type of searches that differed qualitatively. Two types that we understood as ‘symbolic’ control, and one type we understood as ‘instrumental’ control (Holmberg, 2003).

The symbolic kind of control was observed in stationary searches. It was made clear what was going on, by the very visible presence of police, their informative signs, and the presence of the metal detector gate. Despite the controls being conducted in relatively enclosed areas, there were seemingly no obstacles in place to hinder people to turn around if they wanted to avoid being controlled, and officers on site were stationary. Holmberg (2003) showed how Danish officers would control non-suspicious drivers to stop and arrest car thieves, hinder other related crimes, but also to make the general public feel safe. In the same manner, the ‘massive presence’ of police (ibid.: p. 139) could have a deterrent effect on weapon related crimes in our case. A flyer provided to controlled individuals marked the end of the performance, and a rationale for the control. Therefore, we interpret these control as symbolic, with focal points on the legitimacy (i.e., non-biased controlling), public perceptions of safety (i.e., the police are visibly ‘doing something’), and deterrence through visibility (i.e., deterring people from carrying weapons in public space because they may encounter such a control), rather than strictly on findings weapons.

The latter form, however, resembled stop and search-practices to a large extent. Contrasting the stationary searches, these controls were more instrumental and target- and efficiency-oriented in their form. There was less time spent setting up the scene, and we observed the officers moving more rapidly between control spots. They also seemed generally more oriented towards finding weapons by their micro target-selection (i.e., decided on the shift, not by the organization) of people and places. There were less options to escape the to-be-controlled area, as was observed when the police entered a bus to control everyone on it in during one of the observed shifts. In another observation, two teams broke off from the stationary controls to conduct these place-based controls in the example described, strengthens this view –such division of labor ensures both that the control ‘ritual’ is performed whilst also maintaining efficient instrumental controlling.

Then, the statistical analyses revealed that the only significant association between the individual characteristics and the probability of being stopped, was age, and that young people, typically in their late 20s, were searched disproportionaly often. For the gender and ethnicity variables, the analyses combined showed no statistically clear relationship.

Due the pilot being cancelled during data collection, our analysis could not be conducted with a well-powered sample, and as such, the mentioned findings should be interpreted with caution.




Holmberg, L. (2003) Policing Stereotypes: A Qualitative Study of Police Work in Denmark. Glienicke/Berlin ; Madison, Wis: Galda + Wilch.

Jobard, F, and Lévy R. (2011) “Racial Profiling: The Parisian Police Experience.” Canadian Journal of Criminology and Criminal Justice 53(1):87–93. Doi: 10.3138/cjccj.53.1.87.

Lindegaard, M. R., et al. (2023) Citizen profiling during weapon controls in Amsterdam: An observational analysis of practices and potential biases.


“Your Breath, Your Voice” – Stop and Search featured in an Exhibition and about George Floyd, Protest, and Antiracism in Norway

Read previous

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.