Especially in the current Corona crisis, more and more companies are turning to online assessments to select the right talent for their business. Online assessments offer the possibility of carrying out a wide range of tests via the Internet, independent of time and place. The procedures vary from measurements of cognitive abilities to knowledge tests and job-related personality tests.
But the advantages of such time- and location-independent digital solutions are not only evident in the current Corona crisis. Compared to physical selection procedures carried out on site, online assessments are particularly convincing because of their high Objectivity and Economic efficiency (see also: test quality criteria in personnel selection): Companies receive an objective assessment basis of the applicant’s skills and also benefit from high potential for time and cost savings. This is because personal selection days and assessment centers are usually very time-consuming and expensive. And it’s not just companies that save time and money – applicants also benefit from the greater flexibility and save themselves long journeys to applicant days. That sounds like an optimal solution for everybody, who is involved. But is this really true? Do online assessments have the potential to completely replace physical selection processes, even after the Corona crisis?
Even though aspects of cost-effectiveness clearly speak in favor of this, a central weakness of online testing procedures has so far spoken against it: The increased risk of fraud. This is because, unlike on-site selection processes, which are usually conducted under physical supervision, the applicant in a digital test situation is often unobservable. Many HR managers therefore fear that online selection processes could become a paradise for cheaters. After all, applicants are highly motivated to achieve a good test result in order to land their dream job. Then why not research the answers to a knowledge test on the Internet, ask a math student friend for help with a math problem, or consult a native speaker for a language test? The skepticism of many HR managers does not seem to be entirely unfounded.
As a rule, the answer is: No. At least not if some important precautions are taken when designing the testing procedure. This is because as the number of online testing methods grows, so does the number of methods that enable the detection of fraudulent behavior. We have selected some of the most interesting processes that could also be exciting for you.
Overall, two different types of fraud can be distinguished whose probability of occurrence in professional personnel selection must be reduced (see figure below). On the one hand, there is the risk that applicants deliberately manipulate the test result (also known as faking) by using unauthorized aids such as textbooks or sources on the Internet. On the other hand, there is a risk that applicants will ask for help in completing the test or even ask someone else to complete the entire test procedure (also known as impersonation).
To reduce faking, companies can already do a lot in their communications. For instance, it is advisable to announce the test not as a selection test, but instead as a self-assessment to test their own abilities. If applicants believe that the test is primarily about identifying their own strengths and weaknesses, then they are naturally less motivated to cheat. More and more companies also require a statutory declaration of the applicant confirming that the test was completed without the use of unauthorized aids. Similar to the falsification of certificates, cheating in online assessments can also be classified as attempted fraud under employment law. Similarly, warnings that cheating on the test can be detected have been shown to be effective (e.g., Dwight & Donovan, 2003).
The structure of the test procedure can also help to reduce the risk of faking to a minimum. To prevent applicants from using solution patterns, the questions should be presented in random order, for example, and the test content should be varied slightly between applicants, for example, by exchanging the numbers in arithmetic questions (this is also referred to as “permuting test content”). Furthermore, time restrictions can help prevent the use of unauthorized aids. So-called rapid response measures, which often require a response from the applicant in only a few seconds, are particularly suitable for this purpose (Maede et al., 2020). Research information in textbooks or on the Internet? The applicant simply does not have time for this.
In addition to design options in terms of communication and test setup, companies are increasingly benefiting from modern software solutions that can be used to detect fraudulent behavior. For example, a change of browser (e.g., away from the test procedure to an Internet page, cf. “PageFocus”) can be detected and taken as an indication of faking. For data protection reasons, only the change of browser is recognized and stored, but not the page accessed.
It becomes more difficult when it comes to preventing impersonation, i.e. unauthorized assistance by third parties. While many companies are already using procedures to prevent faking profitably, they are still struggling to implement procedures to prevent impersonation. Especially in personnel selection, however, it is of course essential to ensure that the applicant actually takes the test and not someone else. In recent years, scientists have therefore developed various new methods to authenticate applicants in selection processes. These are usually based on the use of a) physical characteristics, b) knowledge bases or (c) behavioral characteristics of applicants.
One way to ensure that the applicant is actually taking the test and not someone else is to record the applicant’s physical characteristics at the beginning of or during the testing process. Two methods, which are significant for personnel selection, are authentication by fingerprint (e.g., fingerprint scanner on smartphone) and facial recognition (e.g., webcam images). In the context of facial recognition, applicants are either watched during the entire test situation (similar to a physical procedure) or images of the applicant are taken at various randomly selected moments. The captured images are then compared with the applicant profile. The advantage of these methods, of course, is that they allow for nearly 100 percent fraud detection-because physical properties are not transferable (Sabbah et al., 2011). The disadvantages, however, include the need to provide the technical equipment (e.g., webcam, fingerprint scanner) and possible data privacy concerns. Storage of the data on secure servers and deletion of all data after completion of the selection process must be guaranteed at all times.
Other methods seek authentication based on a unique knowledge base of the applicant. During the test procedure, he or she is asked questions that are highly unlikely to be answered by a third person. Appropriate questions can be answered by information in application documents (e.g. information in the curriculum vitae) or generated in upstream learning or testing environments (one also talks about Dynamic Question Profiling, Ullah et al., 2019). Research shows that these procedures are very well suited to detect third-party editing (Ullah et al., 2019). The disadvantage of the procedure is that collaborative editing in work groups cannot be completely ruled out.
Another option is to measure behavioral characteristics of the applicant (e.g., keystrokes, mouse movements, behavior, and response patterns) and use these for authentication. For example, a user-specific authentication model can be computed from the applicant’s behavior in a short pre-switched mini-game (Mohamed & Saxena, 2016). Thanks to continuous advances in machine learning, such user profiles can usually be determined in just a few minutes and with high accuracy (e.g., using artificial neural networks or random forest models). The applicant’s behavior during the test procedure is then regularly compared with the user profile created. A significant deviation (determined via equidistant measures) can be considered as an indication for impersonation. Previous research supports the success of this method, with impersonation detection successful in 95-97% of cases, based solely on the data collected in an entertaining mini-game (Mohamed & Saxena, 2016). There is also the possibility to perform the test during a final on-site interview to finally confirm the identity of the applicant. The advantages of these procedures are that they can be carried out in a short time and at low cost, and that they are based on strict data protection guidelines. However, its use requires expertise in the field of machine learning and, consequently, usually the cooperation with a provider who has the corresponding know-how to create appropriate user profiles.
As the brief look at a few of the many existing fraud detection techniques shows, this can be accomplished even in an online-only environment. It is advisable to orient the procedures to the target group, the position to be filled and the test procedures used. You would like to make your selection process fraud-proof and give“cheaters” no more chance? We would be happy to assist you!