After a year of posting I feel like I’ve put off talking about my own research for far too long. My thesis project was part of a set of studies that had been ongoing for a few years before I joined my master’s program in 2009. I was unsure about joining the research lab because my focus was on social psychology while they were investigating eyewitness memory (cognitive psychology). Since the lab used cognitive-behavioral research methods, I could use the experience gained here with just about any other lab. Even if their research was not my main focus, it was still a good opportunity for me to learn about methodology.
I worked with another graduate student who wanted to create a simple intervention to help people improve their eyewitness memory. Basically, participants sat in front of a computer screen where they were shown faces. Each face would appear for three to five seconds (the timing was altered in a few version of our set of studies) and was in the center of the screen. They also see geometric shapes with different fill patterns, and nonsense word pairs. Examples of our stimuli can be seen below.
So far so good right? The participant simply looked at the face, pattern, or word presented and tried to remember it for the test. A simple memory test would mix in items they had never seen with items that were on their study list and their task would be to determine if the item was old (one they studied earlier) or new (one they had never seen). Our test was a bit more complicated.
During the memory-testing phase, the participants were faced with old items , new items, and recombined images (morphed images combining two old items). Here are examples of each stimulus type:
The top two items of each set are images you may see during the study portion of the test. The item below the pair is a recombination of the old items creating a new item. For faces, the outer features (hair, ears, chin) are combined with the inner features (eyes, nose, mouth). For shapes, outside shape is combined with inside fill color. For words, the first word part is combined with the second word part. The resulting image is a recombination (also known as conjunction image).
The participants were tasked with identifying items on the study list as old and all others as new (completely new items and the morphed items). Even though the recombined morphed images were made up of old parts, they were still required to identify them as entirely new items. Different colored fill patterns would be used with different shapes. For words, old prefixes would be added to old suffixes.
There are a few key terms with this kind of test. First, when someone correctly identifies that an image is old or new, we call it a hit. If they call an old item new, this is called a miss. If they call a new item old, however, we call this a false alarm for our study. As you may imagine, misidentification for an eyewitness is a dangerous problem. Thus, we always try to find ways to reduce the rate of false alarms without making people overly skeptical of themselves.
Why else might we use recombined images? Theorists have shown that human visual processing is not a simple matter of remembering a distinct feature about someone’s face (such as having a large nose). Experts have shown that it is the unique way the parts combine together which makes up an experience. Below is the famous Ponzo illusion which you have probably seen. It’s not just for fun; this illusion actually has a practical application that I’m about to explain.
Even when you know the top line is the same length as the bottom one, we still perceive it as being a different length. I cannot stress this enough – your brain will automatically interpret stimuli from the perceptual level, and when that message gets to your working memory, you can finally begin to process the items individually. Your brain is telling you the lines are different before you have the chance to begin thinking otherwise.
This example supports the theory that human visual processing is configural – we perceive items based on their relationships to other items! While we may remember the distinctive features of some faces, such as a person with a large nose, we still only remember it based on context. Perhaps that nose looks too large compared to the other features on the face, but if it were placed on a different person’s face, it may appear to be too small. We remember details based on their context!
A quick note about the patterns and the words – if human visual processing is completely part-based, we should have a good memory with random shapes and words too. However, we know with a high amount of confidence that the brain uses different parts to process faces and other objects. Most objects are processed by a part called the “parahippocampal place area” (brain part near the hippocampus) while faces are processed in the fusiform face area. Some other theorists suggest items that we are exposed to frequently (item expertise) are processed in the fusiform area – it’s not just for faces. It only appears that way because faces are something that most humans are exposed to more than almost any other objects. There is still a bit of debate, but I wanted to show why we used faces, patterns, and words in the study – to help determine if visual processing was configural or part-based.
Now back to the experiment!
As you may have guessed, when a participant saw these recombined morphed images, they had a higher rate of false alarms (they reported the morphed faces as one they had seen). It’s a fairly hard test, so that should be expected. As I mentioned, we did have an intervention in mind.
For the control group, we gave the tests as I had mentioned above, but for the experimental group, we added feedback on their accuracy. After each judgment (was the item old or new), we added a feedback screen that simply stated, “Your last judgment was (correct or incorrect).” The feedback would state your overall accuracy as a percentage and state the type of error (if you made an error). For example, if I saw a recombined image and said it was an OLD item, the feedback screen would say, “Your last judgment was incorrect. Your current accuracy is (whatever % I had) – the last item was a recombined image.”
By using these morphed faces and recombined items we are actually trying to increase the rate of false alarms. Ideally, if we can show that our method reduces the rate of false alarms without reducing the rate of hits, then we have a good intervention
Using this method we gathered our data over three instances of this study. What did we find? Unfortunately, we failed to observe any effect for feedback. This was a bit upsetting, but that’s the scientific process for you. A lot of theories suggest that eyewitness memory is not something you can improve after the memory creation.
At this point, I wanted to add something new to the study. The focus of my master’s program was personality and social psychology. While administering this test, I noticed that some participants seemed more invested in their performance than others. When the feedback told them they made a mistake, they appeared to sigh. When the feedback said they were correct they would sometimes nod or crack a small smile. On the other hand, other participants seemed completely unmoved by the feedback. With my exposure to personality psychology, I theorized that personality factors might make someone more invested in their performance, even on a simple test like this one.
I pitched the idea to the head of the lab, but he thought a broad personality test lacked precision. If we wanted to measure something, it needed to be directly related to what we observed. After a bit of back and forth, we settled on measuring competitiveness. This seemed appropriate as the subjects sat in the same room together, though not in direct competition with one another, and were aware of other people taking the test with them. Furthermore, it seems reasonable that someone with a high level of competitiveness would want to achieve the highest accuracy rating possible.
I searched around for a good way to measure competitiveness and settled on a test devised by Smithers and Houston (1992) as it was shown to be a good predictor in a meta-analysis of measures of competitiveness in 2002 (Houston, McIntire, Kinnie, & Terry). The questionnaire measures competitiveness in a fairly straightforward way while asking questions like “I often try to out perform others” (rate your agreement on a 1-5 scale, 1 being strongly disagree, 5 being strongly agree – see below for more examples). A total of 17 questions measured competitiveness and we decided to place them at the end of the memory test. In most cases you want to randomize the presentation of materials, but in this case we felt that alterations to the testing method for the eyewitness portion of the study should be avoided.
Items marked with an asterisk are reverse coded (agreeing to reversed items means you are less competitive). There are a few more questions that this in the actual tool, but as you can see, the questions are fairly simple.
Over the course of the summer of 2010, I gathered the data for this study. After a bit of digging, we found some interesting results.
We discovered no effect for feedback on pattern or word memory at all. One could conclude that our method had no effect on part-based memory. We observed a small difference between groups for rate of error with recombined (morphed) faces. Participants in the control group appeared to miss them slightly more frequently. Interestingly, the control group’s miss rate was highly correlated with competitiveness. When a highly competitive individual responded, they were more likely to mistakenly identify a morphed face as a face they learned on the study list. In the feedback group, however, the competitive and non-competitive people responded with the same accuracy!
As shown above, we can see a large interaction effect here. The group receiving feedback (the blue line) does not show a relationship between accuracy and competitiveness (hence a flat line showing no correlation), while the control group (the black line) does show a strong relationship (the steep line showing nearly a .70 correlation). So if you are competitive, and you do not get feedback, you have a higher false alarm rate to those recombined faces! Since this was not a test-retest method, I cannot really say that the feedback condition caused a change in behavior, but it seems to be very probable. It would appear that the effects for competitiveness can be completely mediated by our intervention, but more testing would need to occur for us to come to this conclusion.
This was a great grad school experience. Getting to work hands-on with real research was interesting, and getting the chance to test my own hypothesis was quite meaningful. On top of that, the experiment turned out to show a really strong effect. I wish the resources would have been available to replicate the experiment while I was still in school. I would love to see the material published, but my lab did was not focused on personality or social effects. Support came from my lab-mates and professor, but they did not have the same background as me, so continuing the studies after I graduated proved difficult.
Overall, I found the experimental process to be very rewarding. Any time you can show the power of personality psychology in action is a success to me!