Detailed memory in older adults

We often refer to older adults as exhibiting memory impairments, or reduced memory specificity. In fact, it can be difficult to try and convince anyone otherwise. But is this an accurate portrayal? In this blog post I want to discuss the implications of such sweeping statements, and whether they are justified by the available data.

First, some thoughts based on anecdotal evidence.

Older adults themselves often complain that their memory “isn’t what it used to be”, or that they are “forgetting things more and more” as they get older. I’ve lost count of the number of times I’ve had this conversation, both with older adult participants in my studies, and with older adults I know socially. In fact, I’ve had the same conversations many times with people who are much younger – people in their 30s, 40s, and 50s. When I ask them about the type of memory failures they have experienced, people usually tell me that they forget the names of famous people, or that they walk into a room and forget why they went there. These kinds of memory failures happen to everyone; they may become more common as we get older, but they are certainly not restricted to people over a certain age.

I can’t remember a single instance in which a healthy adult, of any age, has told me that they have trouble remembering events and experiences after they have happened.

Ageing memory in the lab

Yet this is exactly what we find when we measure older adults’ autobiographical memories in the lab. We find that their memories are less detailed than those of younger adults (e.g. Levine et al., 2002), or that they tend to recall general memories that are composites of either semantically similar, or temporally proximate, experiences, rather than descriptions of specific one-off events (e.g. Piolino et al., 2006).

This dovetails neatly with the deficits we find when we ask older adults to remember lists of words they learned a few minutes earlier (Clarys et al., 2009; Craik & McDowd, 1987; Nyberg et al., 1996), to remember their way around a virtual environment (Yamamoto & DeGirolamo, 2012), or to remember the associations between pictures that were presented and the locations on the computer screen in which they appeared (Siegel & Castel, 2018).

In short, together these findings suggest that older adults simply don’t remember as well as younger adults.

When stated like this (as it is in much of the published research showing age-related memory deficits), it sounds as though older adults are unable to recall detailed or specific memories.

This is not true.

In most studies of autobiographical memory, participants are asked to recall several different memories, and their scores are derived from averaging what they recalled across the whole set. Yes, on average they may remember fewer details than younger adults, but within the set of memories there are likely to be some that are very well-remembered and some that are not well-remembered at all (more on that in a later post).

The graph below shows the number of episodic (event-specific) memory details recalled (y-axis) by older and younger adults when we asked them to recall a single memory from between the ages of 11 and 17 (the x-axis represents the 20 participants in each group, ranked, within group, in order of lowest to highest episodic memory score).

The black reference line shows the median score for young adults. Older adults certainly seem to recall fewer details, but let’s not forget that their memories are somewhere between 48 and 69 years old. Younger adults’ memories are much younger – between 2 and 19 years old. That’s not a fair test.

This next graph shows the number of episodic details recalled by the same participants (not in the same order), this time when asked to recall a memory from the past year. This time their memories are roughly the same age.

The top-performing young adult remembered more than twice as many details as the top-performing older adult… but also more than twice as many details as the second-best young adult. This person’s data point is an outlier – it is more than 3 standard deviations above the young adult average score. The individual who produced this score was recalling something that was clearly exceptionally memorable. It was NOT the same person who recalled the most details in the memory from age 11-17. Aside from this individual, younger adults don’t look any better than older adults on this task.

Whether we mean to or not, we tend to talk about memory performance as though it is a trait, and we rarely discuss the interaction between a person’s memory ability and other factors – how long ago the event happened, or how inherently memorable it was. It would be absurd to suggest that last Tuesday’s breakfast should be remembered in as much detail as your wedding day, or the birth of your first child. Indeed, as far as I’m aware, no one is suggesting that, but we aren’t really talking about (or formally investigating) these issues either.

If you’re trying to prove that older adults’ memory is impaired, one of the most reliable autobiographical memory tasks is the Autobiographical Memory Test. In this task, participants are given a cue word, and asked to recall a memory associated with that word. They are asked to recall a specific, one-off memory that lasted a day or less. The memories that participants report are scored on a scale of 0 (no memory) through 1 and 2 (memory of a repeated event or an event that lasted longer than 24 hours, with or without detail), to 3 and 4 (memory of a specific event with or without detail). Typically, participants are given a list of words, and attempt to recall a different memory associated with each, and their score is the average specificity rating of all the recalled memories.

We did this with the same participants as above. Young adults scored, on average 3.43 out of 4 (SD=.56), while older adults scored an average of 2.60 (SD=.54). This difference was highly statistically significant. The averages we calculated were across 20 participants in each group, and across 6 memories per participant. According to this analysis, older adults’ autobiographical memory is impaired – again, this suggests they are unable to remember as well as younger adults.

Let’s look at the results another way. Here are the 20 younger and 20 older participants in rank order, with the y-axis representing the number of memories (out of 6) for which they scored 3 or 4 points (i.e. specific memories with or without detail).

It’s clear that a greater number of young adults’ memories score 3 or 4 points, but note that every single participant recalls at least one such memory, and the vast majority of people recall several. Older adults are not unable to recall specific memories – they are perfectly able. They just do so less frequently than younger adults.

The question of why is another matter. I don’t know of any studies that have attempted to understand how participants approach this task. When faced with the requirement to recall any memory, from any time period, that may be only loosely associated with a generic cue word (e.g. happy), how do people engage in the the kind of mental search that will result in the recall of an appropriate memory? We tried some very brief and informal investigations, asking people to think aloud when they were given the cue word – for the word happy, people tried to think of “the happiest I have ever been”, or “the first time I remember being really happy”. Whether this is consistent with what researchers think is happening, I don’t know. It’s also quite feasible that this task could be approached differently by older and younger adults, particularly given that older adults have to search a much larger set of accrued experiences.

What’s clear is that autobiographical memory is more complex than we usually like to admit, and that when we think we are measuring autobiographical memory ability, if we ignore the features of the events we ask them to describe, then we may be doing nothing of the sort.

 

 

Quantifying narrative memory

One of the most frustrating tasks in everyday memory research is the laborious process of transcribing and coding memory details in order to measure how well a participant remembers an event. In this blog post I look at some of the issues coding presents and speculate about what the future of quantifying memory might look like.

Inherent subjectivity

The most obvious problem we encounter when trying to turn qualitative memory narratives into quantitative data is the subjectivity involved in determining what counts as a detail. One of the most influential coding schemes so far was developed by Brian Levine and colleagues (Levine et al., 2002). On the face of it, counting details doesn’t sound all that difficult; written language can be divided pretty neatly into phrases and sentences that each express an individual idea, and in an ideal world spoken language would follow roughly the same rules.

However, spoken language is much more messy than written language. We are, of course, used to listening to speech, and perhaps we filter out some of the uncertainty and redundancy when we hear people tell a story. But seeing free narrative speech transcribed, one begins to realise how difficult the task of coding will be; there are no hard and fast rules.

Look at this quote from Charles Dickens’ Great Expectations:

“I took her hand in mine [1], and we went out of the ruined place [2]; and, as the morning mists had risen long ago when I first left the forge [3], so, the evening mists were rising now [4], and in all the broad expanse of tranquil light they showed to me [5], I saw no shadow of another parting from her [6].”
― Charles Dickens, Great Expectations

I have underlined what, to me, are separate details, and put a number in brackets after each one. Number 6 is metaphorical, so let’s ignore that one.

Numbers 1, 2, 4, and 5 are what we might consider to be internal episodic details – those embedded within the spatiotemporal context of the event that is being described. Detail number 3 is also episodic, since it refers to a specific time and place, but it refers to a time and place other than the one in question, so we might call it an external detail (i.e. external to the main event). We could argue about whether 4 and 5 are really separate details, since the mists rising and the broad expanse of tranquil light both describe related atmospheric conditions, but generally the details in this short passage are quite easy to identify.

Now look at a quote from a participant’s memory of an event that happened two weeks ago:

“We had lunch in Somerset House there [1] and then it was too early to go home [2] so we

…there was another exhibition which was someone’s paintings and they were translated into carpets [3], which was quite good [4] but, I mean, none of the carpets would I actually have in my house [5]. And they were colossally priced [6] because they were hand made, you know, each knot is done [7], which… I love old carpets , hand made ones, that’s a great passion of mine [8], but these were just sort of, erm… There was one which was called ‘hospital doors’ [9], and someone saw these hospital doors and they were sort of square with two round bits like that, two sort of port holes and then some sort of panel here [10] and this was translated on to a carpet [11]. I don’t think it worked at all [12], and they were done in various sort of colour schemes, you know [13], these round bits were a little bit slightly raised somehow [14]. And then we went home [15].”

 

Details 1 and 2, above, are pretty straightforward, but then it begins to get messy. Number 3 could feasibly be divided into two details: going to an exhibition, and paintings translated into carpets. I’ve underlined it as one detail, but if the participant had conveyed the same information in two separate sentences, would I have been more likely to count it as two?

Numbers 4, 5, and 12 could be episodic memory details, if they reflect things the participant recalls thinking at the time (imagine the ideas prefixed with “I remember thinking…”), but they could equally be thoughts generated during the retrieval attempt, merely based on what was remembered about the original event.

Number 8 is a semantic detail – one that is not tied to a specific time and place.

Numbers 9 and 10 are tricky. They could be counted together as one: memory for the particular piece called hospital doors. I counted them as two, one for the name, and the other for what the piece looked like. On the other hand, number 10 could feasibly be divided into sub-details, one for each of the visual components that were recalled… if we skip almost to the end of the quote, to number 14, there is further description of the same piece. Should this be tacked onto the description in number 10, as another visual detail? Possibly. But then, the information clearly wasn’t recalled as one detail, so perhaps not.

These are the kinds of decisions that must be made over and over again during coding, and this is the main reason why the process of coding takes such a long time. It’s extremely difficult to set up a scheme in advance that can absorb all of the uncertainty and ambiguity one is likely to encounter.

So what can we do?

The first thing to do is to decide what the participant is trying to say, which is no easy task. Different people use language in different ways: some people are concise while others are talkative; some people using hedging language (I think, it might have been… etc.) while others talk with much more certainty; some people’s speech is careful and controlled, while others talk quickly, in fragmented speech, and repeat themselves often. It is therefore important to get at the underlying meaning – sometimes it helps if an utterance is reworded as a basic idea unit, as I have done in some of the examples above.

Coarse grain or fine grain?

We also need to consider the grain of the coding. At a fine grain, we might code for individual visual components recalled about a particular object, and at a coarser grain we might code for memory of the object, regardless of the number of details recalled about it.

For example: “I remember seeing a black dog, with long shaggy fur and pointed ears. It had a white patch over one of its eyes“. At a fine grain we might code four details, while at a coarse grain we might just code one (“dog“). In fact, at a very fine grain we might code for more than four details: dog, that is black, long fur, shaggy fur, pointed ears, white patch, over only one eye.

Consider these three possible memory details about the dog:

1. There was something distinctive about its face
2. It had a patch on its face
3. It had a white patch on its face
4. It had a white patch over one eye.

As you go down the list, you can see that these details increase in specificity. It might seem like I’m being pedantic, but in a laboratory memory test a participant would usually be required to recall the precise stimuli they had seen during a study phase. A memory composed entirely of details at the specificity level of number 1, above, would come across as vague, and we might say the participant didn’t remember the event very well.

Generally, the finer the grain, the more difficult the coding becomes. One has to constantly question which details could be recalled independently from the others, and where it might be possible to make errors or to fail to remember something clearly.

Coarse grain coding is a breeze by comparison. At its most extreme, this takes the form of rating an entire memory on a predefined specificity scale, e.g. (from Piolino et al., 2006)

0 = no memory

1 = repeated or extended event with no details

2 = repeated or extended event with details

3 = specific one-off event with no details

4 = specific one-off event with details

The problem with this sort of rating is that it lacks sensitivity at the upper end of the scale – it doesn’t matter whether the participant remembers 4 details or 400 details about an event, they would still get the same maximum score. It works well as a measure of retrieving the fact of a memory, but I would argue that it doesn’t really get at the kind of rich episodic detail we associate with the lay concept of a good memory. Of course, its suitability depends on the aim and design of the experiment.

Dealing with subjectivity

We can deal with the issue of subjectivity, in part, by asking independent coders to count the number of details in a subset of transcripts, then determining the extent to which the original coder and independent coders agree with one another. This gives us a measure of the reliability of the coding system. This is an essential part of the process, and ensures that the coding within a particular study has been carried out consistently. However, it tells us about the correlations between the scores awarded by different individuals, and is less helpful for measuring the true number of details a memory contains. This means that it is difficult to compare the numbers of details across different studies, and even across different memory tasks within the same study (I will write more about this in a separate blog post).

What might the future look like?

I think the future of narrative coding looks automatic. There are programmes like Linguistic Inquiry and Word Count (LIWC) that have been around for years already, which can run powerful text analysis that has been shown to correlate with a number of different psychological constructs (Pennebaker, Mehl, & Niederhoffer, 2003). If we could develop a specialised dictionary for these programmes that could automatically code memory transcripts then we could save ourselves a huge amount of time, and a huge amount of worry over subjectivity. Transcribing and coding memories is by far the most time-consuming part of naturalistic memory research, and can be off-putting to researchers considering working in this area. Automation could be transformative.

 

Levine, B., Svoboda, E., Hay, J. F., Winocur, G., & Moscovitch, M. (2002). Aging and autobiographical memory: Dissociating episodic from semantic retrieval. Psychology and Aging, 17 (4), 677-689.

Pennebaker, J. W., Mehl, M. R., & Niederhoffer, K. G. (2003). Psychological aspects of natural language use: Our words, our selves. Annual Review of Psychology, 54, 547-577.

Piolino, P., Desgranges, B., Clarys, D., Guillery-Girard, B., Taconnat, L., Isingrini, M., & Eustache, F. (2006). Autobiographical memory, autonoetic consciousness, and self-perspective in aging. Psychology and Aging, 21 (3), 510-525.