Up to now, the potential of eye tracking in science as well as in everyday life has not been fully realized because of the high acquisition cost of trackers. Recently, manufacturers have introduced low-cost devices, preparing the way for wider use of this underutilized technology. As soon as scientists show independently of the manufacturers that low-cost devices are accurate enough for application and research, the real advent of eye trackers will have arrived. To facilitate this development, we propose a simple approach for comparing two eye trackers by adopting a method that psychologists have been practicing in diagnostics for decades: correlating constructs to show reliability and validity. In a laboratory study, we ran the newer, low-cost EyeTribe eye tracker and an established SensoMotoric Instruments eye tracker at the same time, positioning one above the other. This design allowed us to directly correlate the eye-tracking metrics of the two devices over time. The experiment was embedded in a research project on memory where 26 participants viewed pictures or words and had to make cognitive judgments afterwards. The outputs of both trackers, that is, the pupil size and point of regard, were highly correlated, as estimated in a mixed effects model. Furthermore, calibration quality explained a substantial amount of individual differences for gaze, but not pupil size. Since data quality is not compromised, we conclude that low-cost eye trackers, in many cases, may be reliable alternatives to established devices.

No abstract

The R package fastpos provides a fast algorithm to estimate the required sample size for a Pearson correlation to stabilize (Schönbrodt and Perugini 2013). The stability approach is an innovative alternative to other means of sample size planning, such as power analysis. Although the approach is young, it has already attracted much interest in the research community. Still, to date, there exists no easy way to use the stability approach because there is no analytical solution and a simulation approach is computationally expensive with a quadratic time complexity. The presented package overcomes this limitation by speeding up the calculation of correlations and achieving linear time complexity. For typical parameters, the theoretical speedup is around a factor of 250, which was empirically confirmed in a comparison with the original implementation corEvol. This speedup allows practitioners to use the stability approach to plan for sample size and theoreticians to further explore the method. Sample size planning with the stability approachSample size planning is one of the most crucial steps before conducting an empirical study. The approach-avoidance conflict lies in the desire for reliable conclusions, but the unwillingness to spend resources for large samples. To balance benefit and cost there exist three more or less established paths: power analysis (e.g. Cohen 1988), accuracy in parameter estimation [AIPE; e.g. Maxwell, Kelley, and Rausch ( 2008)] and interval based accuracy methods (Algina and Olejnik 2003). Recently, a fourth way was introduced: stability (Schönbrodt and Perugini 2013). The general idea of this approach is to determine the sample size at which a certain percentage of studies will fall into an priori specified interval and stay in this interval if the sample size is increased further. For instance, if the population correlation is 0.5, one can define the limits to be 0.4 and 0.6. Given these constraints, what sample size is required to guarantee, with a certain probability (e.g. 90%), that the correlation coefficient will not drop below 0.4 or rise above 0.6 if more participants are added. This sample size is also referred to as the critical point of stability for the specific parameters. The stability approach is promising because it (1) focuses on the effect size instead of significance and (2) is fairly intuitive. Indeed, the interest in the method is growing, evident in more than 1500 citations of the original publication. But a proper software package for the stability approach is still missing.

The passt package is an R implementation of the Probability ASSociator Time (PASS-T) model, an artificial neural network designed to explain how humans make judgments of frequency and duration (Titz & Sedlmeier, 2019). The package was developed with two purposes in mind: (1) to provide a simple way to reproduce simulation results in judgments of frequency and duration as described in Titz & Sedlmeier (2019), and (2) to explore the PASS-T model by allowing users to run their own individual simulations.

No abstract