The picture shows that working memory for simple repetition (Forwards Digit Span, Forward Corsi Blocks) has increased slightly over 43 years, whilst working memory for the more complicated task of items in reverse order (Backwards Digit Span, Backwards Corsi Blocks) has fallen slightly.
Digit span is a bombshell of a test. Despite being very brief and simple to administer, it is a true ratio measure (in the SS Stevens sense, with a real zero) of how many digits you can repeat forwards and backwards. Consider it a test of whether you can remember a brief instruction, at least long enough to repeat it; and whether you can hold items in your short-term memory long enough to manipulate them. It is difficult to function well without those capabilities. The first (digits forwards) is mere repetition, but useful. Repeated instructions serve as a guide to action. The second (digits backwards) is basic data management, which is much more useful, though a more demanding task. It is analogous to being able to reconstruct a disassembled system or device. On a broader front, short term memory is a good test of racial differences in ability, since it has minimal cultural content but identifies processing power bottlenecks.
If we have really been getting smarter over the years, as the Flynn Effect suggests, then those scores should be going up as fast as any other subtest scores. Not so. The result is either No Change (as argued by Gignac) or small improvements in digit forwards and small decrements in digits backwards (as argued by Woodley).
Woodley argues that general ability is falling because of dysgenic effects, but also becoming more specialized, which he calls the Co-Occurrence model. Should it be called “Duller but specialized”?
How do these theories fare in the light of a massive new meta-analysis?
By the way, just in case you think this is some terrible battle, Flynn is no more tied to the Flynn Effect than Woodley is tied to the Woodley Effect. Charles Murray named both those effects, the first to sum up a long series of findings about rising intelligence test scores, starting with Rundquist (1936); the latter in an email exchange with me as I was telling him about the first results of Woodley’s research on Victorian reaction times. Charles Murray wondered how psychology would reconcile these contrary trends: the paper to be described is part of that long-term process.
Having conversations and meals with both Flynn and Woodley for several days in Montreal was a real pleasure. They and the wider group discussed trends in intelligence, including finding Flynn’s alarm about falling competence on Piagetian tasks. In addition, Jim Flynn confirmed to me that he had said in one of his books (he thought in 2012) that no-one was researching genetic causes of racial differences in intelligence “because they feared finding something”. Later I will post about Flynn’s view of the falling scores on Piagetian tasks.
Anyway, back to working memory, or short-term memory as it used to be called. The paper is massive in scope, has more study samples than previous publications on this topic, is extremely large with circa 140,000 subjects and is also a massive confirmation of Woodley’s reworking of Gignac’s data, on a far larger scale. It seems that over the last 43 years we have become able to repeat a bit more but manipulate a bit less. We can echo more, and analyze less.
Of course, there is another way of measuring short term memory, and thus to give pleasure to multiple intelligence believers. You can tap a series of objects in a particular sequence, and then ask the subject to tap them in the same sequence. This can be done forwards and backward, as with digits. This is the Corsi block span (1972) task, which I find difficult to give, let alone to observe and score. Nonetheless, it is a useful measure of spatial short-term memory, which is an important skill. So, one can test short term memory in two different modalities, to see if there is a real general effect, at the central rather than peripheral level.
The Flynn effect for verbal and visuospatial short-term and working memory: A cross-temporal meta-analysis
Peera Wongupparaj, Rangsirat Wongupparaj, Veena Kumaria, Robin G. Morris
This is a very detailed study, based on the real scores, not the scale totals, and corrected for age, sex, and the wealth of the countries of origin, which might affect nutritional status. Women did better on digits, men on blocks.
In a good, wide-ranging introduction to the topic, Wongupparaj and colleagues cover the Flynn Effect and Woodley’s co-occurrence model. Forward spans increase over the four decades of study (digits by r=0.12 and blocks by r= 0.10); backward spans decrease (digits by r=−0.06 and blocks by r=−0.17). The authors sum up their findings thus:
The gain on the verbal and visuospatial STM tasks (i.e. FDS and FCB) and decline on the verbal and visuospatial WM tasks (i.e. BDS and BCB) are consistent with the co-occurrence model (Woodley of Menie & Fernandes, 2015). Based on this model, it is predicted that less g-loaded measures should show arise, whereas more g-loaded measures may show the opposite trend as shown in the studies by Gignac (2015) and Woodley of Menie and Fernandes (2015).
The authors add that the increase on the less g loaded forwards conditions suggests environmental causes including practice effects, while the decrease on the more g loaded backwards conditions suggests dysgenic effects, probably the reduced fertility of brighter persons, but perhaps also an effect of ageing populations. They add:
Visuospatial WM has been found to be a strong predictor of mathematical ability (e.g. Ashkenazi et al., 2013; De Smedt et al., 2009; Hubber, Gilmore, & Cragg, 2014; van der Ven, van der Maas, Straatemeier, & Jansen, 2013). The mathematic scores on the International Mathematical Olympiad (IMO) during 1959–2015 indicate a significant and negative trend between the year of competition and efficiency score (ratio of attained score and all possible score) (n = 54, r = −0.71, p < 0.01; weighted by sample size for each year) (IMO, 2016). Furthermore, the current findings are consistent with many studies that have observed decline on attentionally demand visuospatial tasks (Pietschnig & Gittler, 2015; Woodley of Menie & Fernandes, 2016).
These are alarming findings, particularly to me with my Whig view of history.
Talking about moderator variables, the authors say:
The study made a comparison between developed and developing countries, finding that people in developed countries showed higher performance on all four tests. To date, although there has been no previous meta-analysis investigating country type for working memory, a recent study by our group has indicated that participants in developed countries have significantly higher scores on general cognitive ability tests (Wongupparaj et al., 2015). Health and nutrition issues are the main concerns in many developing countries (Müller & Krawinkel, 2005) and these problems have been found to affect the memory span of children (Jukes, 2005; Miu et al., 2016; Niehaus et al., 2002). Furthermore, many studies on illiterate and unschooled people have demonstrated that low scores on STM and WM tasks were usually observed in illiterate people (Kosmidis, Zafiri, & Politimou, 2011). In all, once all moderators (age, sex, test platform and type of country) were included in the model, the relationships between year of publication and mean scores for STM and WM tasks remain statistically significant. This indicated the robustness of the FE and anti-FE on memory measures.
Of course, a possibility that the authors do not consider is that people in developing countries are brighter, which is why their countries are better developed.
The authors conclude:
In summary, verbal and visuospatial STM shows a gradual rise, supporting the FE for STM as measured by FDS and FCB, whilst verbal and visuospatial WM gradually declines over the past four decades, supporting the anti-FE for WM as indexed by BDS and BCB. Over time, environmental influences might have driven the test score changes for both types of STM but speculatively, dysgenic selection against general intelligence and also possibly age-related cognitive decline could have influenced the declining test score for verbal and visuospatial WM, especially for developed countries. These patterns of the results are in line with the predictions from the co-occurrence model, that is, the FE effect possibly occurs on less g-loaded abilities, whereas the anti-FE effect may concentrate on high g-loaded abilities.
So, Woodley’s “co-occurrence” model gets a strong confirmation.
What next? I think that when discussing historical changes in ability we should restrict ourselves to ratio scales with true zeros. That will keep us better grounded in real changes. In addition, we should look at tasks which have a formal logic to them such as Maths, Science and hypothetico-deductive Piagetian formal operations tasks. Back to Archimedes. Come to that, back to Eratosthenes.