Dear Carolyn, 

Not much can be found on this topic, I do not k now if these papers are of any help to you:

Article Verbal working memory components can be selectively influenc...

Article Age Differences in Memory-Span Errors: Speed or Inhibitory Mechanisms?

Article Effects of Late-Onset Depression and Recovery on Autobiograp...

Article Hemodynamic Changes in the Prefrontal Cortex during Digit Sp...

Thank you so much Beatrice for all this helpful information. Will get into those papers right away!

Carolyn

Hi,

Could you clarify whether it's a digit span test per se that you're interested in or the serial recall task used in a digit-span test? If the latter, most articles on serial recall report and analyse no. correct for each item-position (or serial position) separately. For examples, see:

R. Henson (1998), Cognitive Psychology. (journal name)

Hughes, Marsh, & Jones (2009). Journal of Experimental Psychology: Learning, Memory, & Cognition .

Best

Rob

Thanks so much for your reply, Robert. Its the actual digit span task I'm after. That is, immediate serial recall of random digit sequences, auditory presentation. Will check those two articles, thank you.

Hi, I noticed some people are following this question. So in case its useful to anyone else, I'm including here two other suggestions I received via email:

Elliott, E. M., Cherry, K. E., Brown, J. S., Smitherman, E. A., Jazwinski, S. M., Yu, Q., & Volaufova, J. (2011). Working memory in the oldest-old: evidence from output serial position curves. Memory & cognition, 39(8), 1423-1434. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3588592/ See Figure 3C.

Maylor, E. A., Vousden, J. I., & Brown, G. D. (1999). Adult age differences in short-term memory for serial order: Data and a model. Psychology and Aging, 14(4), 572.

Unfortunately, the data in both papers are presented as figures, not in numerical form, so they can only be estimated estimate (for the first, more recent paper, it might also be possible to contact the authors for raw data).

I waited to reply since I don’t think I am offering you much by the following, but here goes.

Libon’s group at Drexel University employed creative method modifications of the DS that may be worth reviewing (Eppig J et al. Dysexecutive Functioning in Mild Cognitive Impairment: Derailment in Temporal Gradients. J Int Neuropsychol Soc. 2012 Jan; 18(1): 20–28. Published online 2011 Oct 7. doi:  10.1017/S1355617711001238), including manipulating the numbers. For example, contiguous numbers would be strategically placed in a Bwd DS, “5269”, a 4-digit example where the 1st and 3rd digits are contiguous numbers, which presumably influence recall. However, because of primacy and recency effects, they limited this strategy.

Edith Kaplan (Boston Process Approach) always made observations about the type of errors one made, and used the whole DS record to examine strategies used. Reviewing DS raw data, this could include such observations as primacy and recency effects. Competing strategies might also be considered, such as, the inability to “release” a number recalled from the previous set of digits. In any case, for such reasons, the isolation of primacy and recency effects when recalling a sequence of numbers is challenging. However, I am following your thread to learn more. Thanks for the question.

As Rob alludes to, the digit span task is designed to measure the maximum list length for which recall is accurate / correct. As such, serial position effects are not so meaningful, because by definition performance is at ceiling up to span (ok not completely true but as a first approximation this holds)

You can administer individually-titrated span or supra span list lengths and then you are likely to observe serial position curves, but at this point you are investigating serial recall, not span.

This is not a simple configuration environment for which to produce norms. For example, many researchers would anticipate different levels of serial recall accuracy depending on just how supra-span the list is, they might expect primacy or recency to vary with details such as presentation rate, and if you are using a digit span task, then there are constraints such as the size of the stimulus set (i.e. you only have 9 stimuli to play with, and they therefore repeat across trials). In other words, norms might be so study-specific or of such limited generalisability that they aren't really norms at all.

My expectation is that most studies, like those you cite, use control groups for exactly these reasons: to provide some meaningful reference point for the ageing data of interest.

Thank you, Kevin, that Eppig paper is most interesting!

On the question of primacy and recency effects, our interest is in the recency effect, because of its usefulness as a marker of effective phonological short-term memory. In the analysis, we'll be using a measure of the recency effect to asses phonological short term memory in brain-damaged patients. We think it might be a purer measure of this aspect of STM than span length itself. 

Getting a good estimate of the recency effect obviously starts with the healthy controls.

Since our analysis will involve combining different sequence lengths, we'll first code responses to a set number of "normalised" positions using the Machtynger, & Shallice (2009) method:

Machtynger, J., & Shallice, T. (2009). Normalizing serial position analyses: The Proportional Accountability algorithm. Cognitive Neuropsychology, 26(2), 217-222.

The question then becomes how to extract from this data a single "index".  Options include a simple difference score between final position and the middle position (which could be defined by position, or defined as the position where recall was poorest overall). Or a measure of slope which includes several positions.

Thoughts welcome,

Carolyn

Thanks for your reply, John. Obviously, we are seeking data for sequence lengths at or above span length. But most studies that report serial position data will do this (they'd be unlikely to report serial position curves otherwise, because all you'd get is a flat line!).

I think its also fair to say we're measuring ISR, not span. Sorry to mislead there.

"many researchers would anticipate different levels of serial recall accuracy depending on just how supra-span the list is..."

Yes, I'm quite worried about this, A related problem is that the optimal sequence length for measuring serial position effects will vary according to the person's own span. And length itself may be a determinant of the depth of the serial position curve. So there may be a confound here.

Perhaps using the word "norms" in my question was also a bit misleading. We'd just like a nice big control group of a similar age to our patients to establish a good measure of recency and use as a benchmark for calculating a similar measure in our patients. Yes, we'd need a similar rate of delivery and mode of presentation (auditory), but since many of these aspects are standard in the literature, we thought there might be a good chance of a match out there somewhere. Of course ideally, we would gather the data ourselves. But certainly worth seeing what's already there first!

Classical studies by D. Broadbent ( Cambridge) were done on digit span experiments in his book ( I believe the title is:  « Work and Performance» in the 1970s) and extenseively covered the field from a behaviorist perspective

Thank you, Jacinthe, will take a look. I guess while I'm at it, I should take a look at Baddeley's working memory book. Will report back if I find anything.

Hi Carolyn,

I'm not sure of the reasoning behind using recency (for auditory lists) "because of its usefulness as a marker of effective phonological short-term memory". I would say that there is general consensus (and this includes Baddeley's view) that auditory recency falls outside the scope of "phonological short-term memory" (see, e.g., Baddeley's book, 2007; see also Page & Norris, 1998, Psych Review, for an example of the modelling of auditory recency using a parameter that again falls outside of the 'phonological loop'). Instead, there is good evidence that the strong recency for auditory (compared to visual) lists is driven by (non-phonological) perceptual organisation factors (see, e.g., Nicholls & Jones, 2002, JEPLMC).

Perhaps manipulating phonological similarity would be a better way to address your question: whether participants find it more difficult to recall, e.g., "b, d, g, t..." than, e.g., "r, j, s, q..." is, within the working memory framework, the classic test of whether participants are using phonological short-term memory (or how intact that faculty is).

Best wishes,

Rob

Rob, interesting perspective, but not consistent with what we know about deficits. The most striking charcteristic of patients with phonological short-term memory deficits is the absence of the recency effect. There is just a steep decline in accuracy towards the end of the list. The primacy effect is always preserved. I'm familiar with the Page and Norris model, and it cannot explain this.

Sometimes patient evidence can just walk all over evidence from normal participants in this way!

Hi Carolyn,

That is indeed a very interesting observation. I must admit that I haven't checked out your work on this (but certainly will do) but my initial take on that particular observation would be that the deficit in that case may not be anything to do with phonological short-term memory per se. For example, if the deficit is specific to i. recency and ii. auditory lists, then the impairment may reflect a deficit in general acoustic organisation processes that operate at the end-boundary of an auditory sequence; these manifest in a 'phonological' memory task but are not specifically phonological. One way to test this would be to add a suffix (e.g., the word "go") at the end of the list. If healthy controls and patients now exhibit a similar serial position curve, then the deficit shown by the latter group in the non-suffix condition is indeed an end-boundary acoustic effect (at least from my theoretical perspective!). Another prediction I would make is that the patients would show the same impairment in a non-verbal short-term memory task (e.g., recall of tone sequences), again suggesting that it's not specifically phonological; have you done anything that might speak to that? (I've just started looking through your publications but if there's a link to something particularly relevant, please let me know).

Best,

Rob

Rob, I don't specialise in the STM area; my area is language disorders. But I've attached a recent draft of a chapter I've written on the topic in case you want a quick overview of the most important findings (only a small portion is about verbal STM).

For example, if the deficit is specific to i. recency and ii. auditory lists, then the impairment may reflect a deficit in general acoustic organisation processes that operate at the end-boundary of an auditory sequence; these manifest in a 'phonological' memory task but are not specifically phonological.

No, the deficit is not specific to items at the ends of lists (although these are disproportionately affected), nor is it specific to auditory lists (although we see less of a decrement in performance on visual lists; see chapter for details). Patients can be entirely normal at nonverbal auditory sequences, such as tones and sounds from the environment. The deficit is often accompanied by other features, such as poor nonword repetition, the production of phonological errors in speech and/or difficulties with phoneme discrimination tasks. A bit awkward to explain all in terms of some sort of auditory end effect, but all entirely compatible with a mild phonological processing deficit that impacts especially greatly on tasks relying heavily on the retention of phonological information.

Recent neuroscience approaches tend to conceptualise STM (phonological or whatever) as an emergent property of the language processing system itself. More Cowan than Baddeley, if you like. Anyway, there's more in the chapter if you're interested.

Thanks for the interesting discussion.

Carolyn

I've been looking for just a histogram or frequency table or raw data set of digit span performance in adults, and it's surprisingly hard to find anything at all! Almost every source reports means or grouped frequencies. Here's a data set that MIGHT be helpful both for my purpose and the old question here from Carolyn Wilshire : Article Benchmarks for Models of Short-Term and Working Memory

I hope this can help:

Jordi Peña-Casanova, Sonia Quiñones-U ´bedab, Marı ´a Quintana-Apariciob, Miquel Aguilarc, Dolors Badenesc, Jose ´ Luis Molinuevod, Laura Tornerd, Alfredo Roblese, Marı ´a Sagrario Barquerof,†, Clara Villanuevaf, Carmen Antu ´nezg, Carlos Martı ´nez-Parrah, Anna Frank-Garcı ´ai, Azucena Sanzi, Manuel Ferna ´ndezj, Vero ´nica Alfonsok, Josep M. Solk, Rafael Blesal, for the NEURONORMA Study Team

Spanish Multicenter Normative Studies (NEURONORMA Project): Norms for Verbal Span, Visuospatial Span, Letter and Number Sequencing, Trail Making Test, and Symbol Digit Modalities Test

Archives of Clinical Neuropsychology Advance Access published August 5, 2009

I have normative data for Auditory, Visual and Visual Apprehension Span by grade level from grades 1 through adult for Letters and Words included is the Errors of Commission. I used letters and words rather than numbers because I wanted to apply it to the problem of reading in children and adults.

Bloomer, R. H. .(2017)Diagnostic Neuropsychological Assessments Vol 2: Individual Short Term Memory Amazon

Helpful thread, thanks for all the answers. Researching this topic I found another useful study which has appeared since the original question was asked.

Choi et al. (2017). A Normative Study of the Digit Span in an Educationally Diverse Elderly Population. Psychiatry Investigation, doi: 10.4306/pi.2014.11.1.39