Schematic Memory: How Expectations Can Hurt

Lindsay Beyerstein of Majikthise asks, "Why do expectations always hurt democrats?" I'm going to hold off on trying to answer that question, and use it as a launching point for a relatively brief explanation of how we should view "frames," or as I'm going to call them from this point on in this post, "schemas"¹. So, to start off, we should define schemas. Schemas are structured representations of objects, events, individuals, social roles, or any other type of concept. In this context, "structured representations"² are just representations that contain both objects and their predicates (relations between objects, attributes, etc.), where relations can be causal, functional, thematic, social, etc. Schemas may or may not be abstract (I don't think they are, but this is an open question), but either way, they represent the gist of several instances of the same concept. Each time we encounter a new instance of a concept, it is incorporated into the schema. Schemas determine what we remember about an instance of a concept, for how long we remember it, and how available the information is. Schemas also provide expectations, so that when we encounter a new instance of a concept, we don't have to process all of the information, but can instead rely on the information already contained in the schema to do most of the work.

To bring all of this schema talk back down to Earth, I'll give an example that's often used in the literature (including by Lakoff, in Women, Fire, and Dangerous Things). Most of us have a lot of experience eating at restaurants. The gist of the restaurant experience is pretty much the same for similar restaurants, and as a result, we have a restaurant schema for these restaurants that contains the general objects (myself, the server, the cook, etc.) and relations (when to order, when the food arrives, when we're given the check, etc.). When we go into a restaurant, we expect things to happen a certain way (e.g., the server brings our food before giving us the check), based on the schema.

Where schemas get interesting is in how they affect our memories and perceptions of instances. There are several interesting memory phenomena related to schemas. The first is that we tend to remember information relevant to our schemas better than information that is irrelevant to them. So, when we encounter an instance of a concept, information that is not relevant to its schema is unlikely to be remembered. In the restaurant example, for instance, we are unlikely to remember information about what the person at the table next to us is eating, because its simply not relevant to our restaurant schema. Just because we don't remember this information doesn't mean that we don't encode it. We do encode much of it, and it is available, at least for a short period of time (a few hours, maybe even a few days or longer), but unless we activate a schema in which it is relevant, we are unlikely to recall it.

Another interesting aspect of schematic memory is that, after a short period of time, we tend to forget which aspects of our schema were present in an instance, and which weren't. In a famous experiment by Sulin and Dooling³, participants were presented with biographical stories. The main character in the story was either given a famous name (e.g., Helen Keller) or an unfamiliar name (Carol Harris). After a delay, participants were given a list of sentences, and asked to determine whether the sentences were from the original stories or not. Some of the sentences they were given were true of the famous person (Keller), but not present in the story. Participants who had been given the story with the famous name were almost as likely to remember having read the new sentences that were true of the famous person as they were to remember the sentences that had actually been in the story. The explanation for this result, which has been borne out by subsequent research, is that people have a Helen Keller schema, which is activated when they read the story with her name. Later, they are unable to distinguish between information that was in the schema but not the story and information that was actually in the story, because both types of information were active at the time of encoding. In sum, then, because instances of a concept activate a schema while those instances are being encoded, and because information from those instances is integrated into the schema rather than being remembered individually, when trying to recall the instance, people often mistakenly remember information from the schema that was not present in the instance.

A third intersting schematic memory phenomena relates to information that is relevant to the schema, but inconsistent with it. For instance, imagine that the server did give us the check before bringing our food. Because our schema provides us with certain expectations about an instance, in this case a restaurant visit, such deviations from expectations will be easily noticed. They will also be easily remembered. If the server had stuck to the script, and given you the check after you finished eating (or just as you were finishing), it's unlikely you would remember anything about it after a few days. However, if the server gives you the check before you get the food, you will probably remember the event for some time to come.

To sum all of that up, schemas make it easier to remember/access schema-relevant information, and in particular, schema-relevant information that is inconsistent with the schema. In addition, we are likely to mistakenly recall information from the schema that was not present in an instance, because the schema and the instance are both active during encoding. There's a lot more to schemas than this (related to social roles, stereotypes, and much much more), some of which I may get to in a future post about how schemas work, but for now, this much will do. The reason I presented this stuff in such excrutiating detail is that I want to describe the way schemas activated by political discourse affect expectations.

At this point, it should be pretty obvious how schemas affect expectations. When we associate a candidate (say Kerry) with a schema (good debator), the schema will license certain inferences, or produce certain expectations about that candidates debate performance. Because of the way schemas work, these expectations will affect what we are likely to remember about any particular debate involving the candidate. If we believe that Kerry is a good debator, then we will expect him to perform well, and any deviation from this will be easily remembered. If he does perform well, we will remember that, but because the debate will be intengrated into our schema, we're unlikely to remember any of the details of the actual debate. Instead, we'll just remember that Kerry was a good debator. Thus, it's a good idea for Republicans to continually associate the good debator schema with Kerry, because 1.) we'll be more likely to notice when he performs badly, and 2.) we're less likely to remember specific good things he did. This is why expectations so often hurt Democrats. In turn, it's also why the help Republicans. We have an "idiot" schema associate with Bush, so if he performs poorly, we won't remember the details, but if he does well, and thus his performance is inconsistent with our "idiot" schema, we will be more likely to remember what he did well.

There are ways to counter this, which draw on the ways in which schemas affect memory. One way is to activate an alterantive schema (Kerry is a bad debator, e.g.). Another is to highlight alternative inferences or expectations from the schema. This is what Lindsay suggests when she writes:

Let's frame this another way: If everyone expects John Edwards to wipe the floor with Dick Cheney (including Cheney and Edwards), then Cheney's at a huge disadvantage. The public will see him as an outclassed bumbler because that's exactly what they expect to see. They're tuning in to see the Golden Boy slay the atherosclerotic dinosaur. If that's what they want to see, that's what they'll see. Cheney is fighting against low expectations, not benefiting from them.

What this involves is highlighting the fact that Kerry (or Edwards, in Lindsay's example) is a better debator because Kerry (or Cheney) is much more intelligent than Bush (or Cheney), and this will not only make him a better debator, but a better leader. When Kerry defeats Bush in a debate, or in a discussion of a particular issue, this shows that Kerry is the right choice, and Bush is a fool whom we cannot count on to take the country in the right direction. This strategy involves activating the same schemas (Kerry - good debator, Bush - idiot), but highilight certain aspects of the schemas so that they are more likely to be active in peoples' minds during the debate. Thus, when they view the debate, and it is consistent with their expectations, they will conceptualize it in a way that causes the expectations to benefit the Democrats. There's also an added benefit: because people have a schema that is good for Democrats active during the debate, they are more likely to mistakenly remember positive things later, even if they weren't present. As in the Sulin and Dooling experiment, the good debator schema will be active while people are encoding the debates, and they will have a hard time distinguishing between information from the schema and information from the debate itself later. Of course, all of this hinges on Kerry actually performing up to expectations, but if the first debate is any indication, this shouldn't be too much of a problem.

1 I'm going to call them schemas, because in the literature that I'm going to be drawing from, that's what they're called. Based on the way I see "frames," and the way they have been discussed in sociology, "schemas" and "frames" are interchangeable. This may not be true of Lakoff's version of "frames," but part of my point is that Lakoff's version of "frames" lacks important properties, possessed by "schemas," that would make them more useful.

² This is a nice graphic representation of a simple structured representation. The directed graph and predicate calculus representations were stolen, and I mean stolen, from Larkey, L. and Love, B. 2003. CAB: Connectionist Analogy Builder. Cognitive Science, 27, 781-794. Click on the figure for a larger view (and an easier-to-read caption).




³ Sulin, R. & Dooling, D. (1974). Intrusion of a thematic idea in retention of prose. Journal of Experimental Psychology, 103, 255-262.