3.4. Programming Experiments in Python with Psychopy¶
|Sassenhagen, J. (2019). Programming Experiments in Python with Psychopy. In S.E.P. Boettcher, D. Draschkow, J. Sassenhagen & M. Schultze (Eds.). Scientific Methods for Open Behavioral, Social and Cognitive Sciences. https://doi.org/10.17605/OSF.IO/
3.4.1. Story Time: Galton’s Pendulum¶
In the 1880s, scientific progress had opened up a novel entertainment option for Her Majesty Queen Victoria’s subjects: have their psychometric traits measured in Francis Galton’s Science Galleries, at the South Kensington Museum. Visitors paid “3 pence” (presumably some sort of currency) and then partake in a series of encounters with novel and strange machinery. Thousands of participants volunteered, establishing one of the first large-sample psychometric databases in history. Amongst the things measured was how long it took the curious museum goer to press a button in response to a simple light stimulus.
Galton measured response times using an ingenious design called the Pendulum chronograph. Simplified, the position of a swinging pendulum at the time point of a key-press could be compared to a reference table. This Galton used as an estimate of response time, which allowed a state of the art resolution of 1/100th of a second. Never before had mental process been measured with higher precision.
Other experimenters around this time were stopping response latencies with manual stopwatches. Galton himself conducted human geographic surveys by walking around London, mentally rating the women he saw for their attractiveness, and covertly noting the ratings with a self made device: a paper cross, into which he, in his pocket, punched a hole indicating if the woman was of below average, average, or above average “beauty”.
It is not advised to mention such practices in a contemporary presentation of scientific results. At the very least, the practices would be seen as grossly inaccurate, and for many, highly open to experimenter bias. In particular, they lack precision. Also, to be frank – they sound like a lot of work! While some might decry a lack of entrepreneurial spirit, most of us will readily admit the advantage of computerized methods.
Progress has made the work of the researcher much simpler. (Remember this sentence whenever you feel frustrated attempting to program an experiment! At least you are not Galton, keeping a pendulum chronograph in working condition, or walking across London to mentally file people into rather crude and subjective boxes.) Today, computers can show various visual input, play all kinds of sounds, and accurately measure … well, a narrow kind of behavioral responses made by experimental participants. Our computer’s graphics and sound cards, and the keyboard and mouse drivers, are relatively arcane. There is as of yet no convenient way to get a computer to show words or pictures on a screen for the purposes of psychological experiments from our favourite programming language, R. There are many commercial, closed-source solutions, which we will all ignore in favour of the powerful and open Python options.
Why? Perhaps the most important benefit computerized experiments have is that they are much more reproducible. Using an Open Source program and making experimental stimuli and scripts available online allows other researchers to 1. exactly retrace what happened in the original experiment, 2. repeat it ad libitum. To actually exploit this reproducibility potential, we must use software that is open. The biggest open source experimental presentation software is Psychopy [Pei07].
3.4.2. Programming Experiments in Python with Psychopy¶
Psychopy allows us to write simple and readable Python code to control our computer’s low-level capacity for displaying and playing stimuli. Why is this necessary? Because we need to work with the computer on a low level in order to get it to achieve highly precise timings, and smoothly display even complex visual stimuli. That is one half of the experimental program; the other will consist in translating the (experimental design) into computer code, so that, e.g., a study participant is presented with the required number of trials resulting from your (power calculation) for the conditions resulting from your (latin square design).
Because Psychopy is written in Python, we having already learned Python, learning Psychopy reduces to learning the Psychopy-specific modules.
The basic logic of experiments in Psychopy¶
Psychopy can display auditory and visual stimuli; visual stimuli may be static,
or dynamic (moving animations, or videos). To display visual stimuli, Psychopy
must know about at least one
Monitor, and at least one
Window. Such a
Window is the plane on which drawn stimuli will be shown. Note that
Window are software objects primarily inside of Psychopy.
They allow Psychopy to show things in a window on your physical screen
Internally, Psychopy knows the backside and the front side of each Window`.
Window is newly created, both sides will be empty. We can now
draw things on the backside (using the
draw method of various visual
objects). Once everything we want to show has been drawn, the
“flipped” so that the painted backside is now shown on the physical screen.
The new backside is blank. We can now draw other things on this blank backside.
One option might be to draw nothing, to show a blank screen after the current
Eventually, we flip again; this clears everything we had drawn on the original
backside, and shows the other side on the screen. So: we manually paint –
piece by piece – one side of the window with everything we want to show, flip
it to show it, paint the new backside, flip again to show and clear, and repeat.
— put image tbd by aylin here —
The visual stimuli we can paint on screens live in the
submodule. This includes various geometric shapes, as well as the
ImageStim classes, which we will discuss extensively in the following.
For movie stimuli, see the MovieStim class. Other stimuli include random dot motion and grating stimuli.
Keeping track of time and responses¶
Psychopy allows collecting button or keyboard responses and mouse events.
For time tracking, one can create and use one or more
To time the duration of an event or interval, the clock is reset before the
event/at the start of the interval, and then measured at the end.
For example, to measure a response time to a stimulus, the clock is reset
exactly when the window flip to show the stimulus happens. Then, the clock
is checked when the button press happens. No pendulums involved!
Keyboard responses are measured by the
functions. If provided with a clock, they return a Python
Storing results and experimental logic¶
Psychopy provides expensive functionality for logging results and the logistics
of presenting stimuli, but it is not even required to learn these; basic Python
code can be sufficient. For example, the humble
A Caveat on Accuracy and Precision¶
In principle, Psychopy can be highly accurate. In practice, much depends on specifics of the experiment and context [GV14][Pla16]. Consider: one study has reported that Galton observed slightly faster response times in Victorian times than are observed in contemporary experiments [WTNM13]. Could it be that the Victorians were mentally faster than us? An alternative suggestion for this has been that timings on digital devices are only ever approximations; i.e., many digital devices could not record increments shorter than 100 ms! Even with modern computer technology, the accuracy of stimulus presentation timing is never better than the screen refresh rate. For example, many laptop monitors have refresh rates of 60 Hz. That is, they can at most show a new stimulus 16.5 ms after the previous stimulus, and all stimulus timing intervals will at best be multiples of 16.5.
Remember the distinction between accuracy and precision: some of the inaccuracy of stimulus and response time collection will be random jitter. In many cases, this will simply show up as noise in the data (and thus, decrease the power of the experiment). Systematic distortions are not a necessary consequence [VG16]. But other aspects represent an inherent bias. For example, for build-in sound cards, auditory stimulus presentation onset is preceded by a delay. Typically, this delay will be approximately the same on every trial; but it will lead to a systematic underestimation of stimulus onsets.
For experiments requiring extremely precise measurements, it becomes crucial to measure, minimize and account for inaccuracy and bias. For this, external hardware is required; i.e., light- or sound pressure sensitive detectors. (For a cheap solution, the Raspberry Pi mini-computer can easily be extended for this purpose.)
An example experiment¶
The following section will guide through the programming of a basic experimental paradigm (a false-memory experiment). It will demonstrate Psychopy functionality required to conduct a typical response time or many other types of experiments. The example will be far from the only way to achieve this goal; many other paths are viable. But following it will show many solutions to typical problems during the creation of a psychological experiment.
3.4.3. Alternative software¶
Another powerful option is OpenSesame [MathotST12], programmed by Sebastiaan Mathôt. OpenSesame provides a graphical front-end, but also allows directly injecting Python code for fine-tuning. It is recommended for those who prefer a point- and-click, mouse-based approach while still demanding an open-source, reproducible tool.
Going online: surveys on the internet¶
Online Experiments with the Psychopy Builder¶
This option is in fact build into Psychopy, but is not available from the Coder view requires for Python programming. Instead, it must be accessed from the Builder interface. See the Psychopy website for a demonstration of how this functionality can be used.
|Pablo Garaizar and Miguel A Vadillo. Accuracy and precision of visual stimulus timing in psychopy: no timing errors in standard usage. PloS one, 9(11):e112033, 2014.
|Sebastiaan Mathôt, Daniel Schreij, and Jan Theeuwes. Opensesame: an open-source, graphical experiment builder for the social sciences. Behavior research methods, 44(2):314–324, 2012.
|Jonathan W Peirce. Psychopy—psychophysics software in python. Journal of neuroscience methods, 162(1-2):8–13, 2007.
|Richard R. Plant. A reminder on millisecond timing accuracy and potential replication failure in computer-based psychology experiments: an open letter. Behavior Research Methods, 48(1):408–411, Mar 2016. URL: https://doi.org/10.3758/s13428-015-0577-0, doi:10.3758/s13428-015-0577-0.
|Miguel A. Vadillo and Pablo Garaizar. The effect of noise-induced variance on parameter recovery from reaction times. BMC Bioinformatics, 17(1):147, Mar 2016. URL: https://doi.org/10.1186/s12859-016-0993-x, doi:10.1186/s12859-016-0993-x.
|Michael A Woodley, Jan Te Nijenhuis, and Raegan Murphy. Were the victorians cleverer than us? the decline in general intelligence estimated from a meta-analysis of the slowing of simple reaction time. Intelligence, 41(6):843–850, 2013.