There are three temporal patterns that emerge from the Agbayani (2013) data:
- For any given calendar year, the age-related pattern of raw intelligence scores shows a peak for test takers in early adulthood, followed by a gradual yet significant decline for test takers at increasingly older ages.
- For any given age category, raw intelligence scores show a consistent increase across calendar years, with the rate of increase being similar for all age categories.
- The intelligence profile for any given birth cohort (which would be obtained by reading diagonally across the grid) shows a relatively flat level of scores from early adulthood until around age 70 or so. This is consistent with most longitudinal intelligence studies applied to individuals.
| Age | Chart A Comparable Intelligence Scores by Calendar Year and Age | |||
| 60 | 80 | 88 | 97 | |
| 40 | 90 | 99 | 109 | |
| 20 | 100 | 110 | 121 | |
| 1960 | 1980 | 2000 | ||
| Calendar Year | ||||
Note that this chart is consistent with the three temporal patterns listed, and in particular note that for the cohort born in 1940 (age 20 in 1960, age 40 in 1980, age 60 in 2000), the raw intelligence profile is relatively flat across adulthood, just as Agbayani (2013) highlights.
In Dickinson (2010), Agbayani (2011) and Agbayani (2013), these results are interpreted in the following way (I will call this the Agbayani interpretation):
- The Flynn effect is a function of cohorts.
- In the absence of a Flynn effect, the cross-sectional age-related declines in intelligence scores will mostly disappear.
That interpretation is almost certainly wrong.
An alternative interpretation to the Agbayani interpretation, one that I find far more natural and plausible, can be outlined as follows (I will call this the Griswold interpretation):
- The Flynn effect is a continuous, universal function of time (not of cohorts).
- In the absence of a Flynn effect, the age-related decline in intelligence scores will emerge in each cohort, demonstrating that such decline is in fact the natural age-related intelligence profile.
To compare the Agbayani and Griswold interpretations and assess their consequences and resulting plausibility, we need to examine each interpretation under a scenario in which there is no Flynn effect, then follow this with a description of how each interpretation can be transitioned into a scenario in which the Flynn effect has full impact. Such an investigation will highlight the fundamental contrast between the two interpretations and will demonstrate how each necessitates a dramatically different understanding of how the Flynn effect must work.
We can begin with the Agbayani interpretation. Assuming a scenario in which there is no Flynn effect and given the Agbayani assumption that there will be little age-related decline in intelligence scores in the absence of a Flynn effect, the resulting two-dimensional grid of raw intelligence scores would now have to look something like Chart B:
| Age | Chart B Agbayani interpretation, no Flynn effect | |||
| 60 | 97 | 97 | 97 | |
| 40 | 99 | 99 | 99 | |
| 20 | 100 | 100 | 100 | |
| 1960 | 1980 | 2000 | ||
| Calendar Year | ||||
Chart B meets the requirements of the Agbayani interpretation. Note that there is practically no age-related decline either at any calendar year or for any cohort, and of course there is also no telltale sign of a Flynn effect. From this base, the trick now is to figure out how to introduce a Flynn effect — that is, transition from Chart B to Chart A — and do it in a way that is consistent with the Agbayani interpretation.
When I first contemplated the Agbayani interpretation, I could see no reasonable way to make that transition work, but upon further reflection I recognize I was being too hasty. In truth, there is a way to make the math work out. What needs to be done is to add something to each birth cohort — call it a Flynn effect boost (FEB) — and in order to account for the fact that the Flynn effect is evident across all age groups (including child age groups), it is necessary to add this boost right from the very beginning (at birth, if you will) and then have its influence remain constant over the cohort's lifetime. Mathematically, the technique looks something like Chart C:
| Age | Chart C Agbayani interpretation, Flynn effect transition | |||
| 60 | 97 x (1 - 2 x FEB) | 97 x (1 - 1 x FEB) | 97 x (1 + 0 x FEB) | |
| 40 | 99 x (1 - 1 x FEB) | 99 x (1 + 0 x FEB) | 99 x (1 + 1 x FEB) | |
| 20 | 100 x (1 + 0 x FEB) | 100 x (1 + 1 x FEB) | 100 x (1 + 2 x FEB) | |
| 1960 | 1980 | 2000 | ||
| Calendar Year | ||||
Chart C starts from the no Flynn effect base of Chart B, then each successive cohort is given a Flynn effect boost that is larger than the boost provided to the previous cohort. Furthermore, each cohort is given the entirety of its Flynn effect boost right from the beginning, with the influence of the boost remaining constant thereafter over the cohort's existence. Reading diagonally across Chart C and observing how the Flynn effect is introduced and maintained in each cohort, we see that this technique is exactly what is required in order to remain consistent with the Agbayani assumption that the Flynn effect is purely a function of cohorts.
If we plug in an FEB value of approximately 0.1, then Chart B transitions quite smoothly into Chart A, thereby demonstrating that the suggested technique provides a plausible mechanism for how the Flynn effect must work under the Agbayani interpretation. I leave open the possibility that I have overlooked something, but as far as I can tell, this technique is the only one mathematically plausible under the conditions of the Agbayani interpretation.
By contrast, the scenarios and transitions look quite different under the Griswold interpretation. We begin once again by considering the scenario in which there is no Flynn effect, along with the Griswold assumption that the age-related intelligence decline will continue to be evident under such a scenario. This means that the two-dimensional grid of raw intelligence scores must now look something like Chart D:
| Age | Chart D Griswold interpretation, no Flynn effect | |||
| 60 | 80 | 80 | 80 | |
| 40 | 90 | 90 | 90 | |
| 20 | 100 | 100 | 100 | |
| 1960 | 1980 | 2000 | ||
| Calendar Year | ||||
Chart D meets the requirements of the Griswold interpretation. Note that the age-related decline is still evident for all calendar years, and furthermore the age-related decline is now evident also for each cohort. Plus there is no hint of a Flynn effect anywhere in these numbers. As before, the next step is to introduce a Flynn effect — that is, transition from Chart D to Chart A — and do it in a way that is consistent with the Griswold interpretation.
Here, the math is fairly straightforward and practically suggests itself. Consistent with the Griswold assumption that the Flynn effect produces a continuous and universal impact over time, the transition is produced simply by introducing a Flynn effect boost (FEB) at each calendar year and across all age groups. Mathematically, the technique looks something like Chart E:
| Age | Chart E Griswold interpretation, Flynn effect transition | |||
| 60 | 80 x (1 + 0 x FEB) | 80 x (1 + 1 x FEB) | 80 x (1 + 2 x FEB) | |
| 40 | 90 x (1 + 0 x FEB) | 90 x (1 + 1 x FEB) | 90 x (1 + 2 x FEB) | |
| 20 | 100 x (1 + 0 x FEB) | 100 x (1 + 1 x FEB) | 100 x (1 + 2 x FEB) | |
| 1960 | 1980 | 2000 | ||
| Calendar Year | ||||
Chart E starts from the no Flynn effect base of Chart D, then a Flynn effect boost is introduced continuously and universally over time, impacting all age groups and all cohorts the same, exactly as required under the assumptions of the Griswold interpretation. If we plug in an FEB value of approximately 0.1, then Chart D transitions quite smoothly into Chart A, demonstrating that the suggested technique provides a plausible mechanism for how the Flynn effect must work under the Griswold interpretation.
Since each interpretation appears to be mathematically plausible, we need to search further for evidence to rule out either (or both). The ideal approach to this task would be to examine an adult population for which there is no Flynn effect: if the cross-sectional age-related intelligence scores are flat over such a population, the Griswold interpretation could be ruled out, and if the cross-sectional age-related intelligence scores are in decline over that population, the Agbayani interpretation could be ruled out. Unfortunately, it is unclear whether any extant human population can be characterized as untouched by the Flynn effect (the Flynn effect's relentless ubiquitousness being one of its more tantalizing features), and thus the ideal approach appears to be unavailable.
If the Flynn effect were to come to a halt, then that would also provide a means for assessing the two interpretations, because the halting would produce distinctly different signatures under each interpretation. Under the Griswold interpretation, a stop in the Flynn effect would immediately impact all age groups and all cohorts, something like Chart F:
| Age | Chart F Griswold interpretation, halt of the Flynn effect at year 2000 | ||||||
| 60 | 80 | 88 | 97 | 97 | 97 | 97 | |
| 40 | 90 | 99 | 109 | 109 | 109 | 109 | |
| 20 | 100 | 110 | 121 | 121 | 121 | 121 | |
| 1960 | 1980 | 2000 | 2020 | 2040 | 2060 | ||
| Calendar Year | |||||||
Under the Agbayani interpretation, a halt to the Flynn effect would produce a more complex pattern, because the halt would impact only future cohorts and not any existing cohorts. This would produce something like Chart G:
| Age | Chart G Agbayani interpretation, halt of the Flynn effect at year 2000 | ||||||
| 60 | 80 | 88 | 97 | 107 | 118 | 128 | |
| 40 | 90 | 99 | 109 | 120 | 130 | 130 | |
| 20 | 100 | 110 | 121 | 131 | 131 | 131 | |
| 1960 | 1980 | 2000 | 2020 | 2040 | 2060 | ||
| Calendar Year | |||||||
Under the Agbayani interpretation, for a period of time after the halt of the Flynn effect, intelligence scores would continue to increase in the older age categories but would begin to level off at the younger age categories (including childhood ages). Such a signature would be strong evidence against the Griswold interpretation and in favor of the Agbayani interpretation. Unfortunately, it is again unclear whether in the real world a halt to the Flynn effect is imminent or whether it could be quickly and easily recognized; and thus this technique, while theoretically interesting, would appear to be pragmatically out of reach.
Despite these obstacles, I do believe some strong arguments can be made against the Agbayani interpretation. The first difficulty is with the assumption that raw intelligence abilities would remain relatively level across the adult years in the absence of a Flynn effect. Although this is certainly possible, it runs counter to many other known biological abilities, such as athleticism and sexual vitality, and it would seem that a pattern of peak during early adulthood followed by gradual decline in later years would be the preferred assumption. To assume otherwise should require some evidentiary explanation, and of course Agbayani (2013) does not provide that explanation — it is only the Agbayani interpretation that supports the assumption of a flat level of intelligence across adulthood, and we have seen that there is at least one alternative interpretation that runs exactly counter to such an assumption.
More problematic still is the need for a Flynn effect boost to be provided to each cohort quite early in its existence, a need driven by the assumption that the Flynn effect is a function of cohorts. Although such a mechanism is theoretically possible, it runs counter to almost everything that is commonly understood about intelligence and the Flynn effect. For instance, you cannot say something like better education might produce the Flynn effect, because under the Agbayani interpretation a cohort's Flynn effect boost has to be fully in place before even the first day of school. Many similar intelligence explanations are ruled out for the exact same reason. By making the Flynn effect a function of cohorts, one removes the element of time, and much of what we understand about the acquisition of intelligence depends upon the passage of time, not upon the introduction of cohorts. Under the Agbayani interpretation, it would appear we must look only for Flynn effect causal candidates that are materially different across cohorts, present in nearly every member of each cohort, and present essentially right from birth. The list of plausible such candidates would seem to be conspicuously small. Although it is extremely common and popular to assume that the Flynn effect must be a cohort-driven phenomenon (see for instance the first sentence here), I think the proponents of that assumption fail to appreciate the difficult-to-explain consequences that must inevitably arise.
Admitting freely to my bias, it seems to me that the Griswold interpretation suffers from far less strain. In the first place it adopts the more physically natural assumption that unaided by a Flynn effect, human intelligence would peak in early adulthood then gradually decline towards old age — mirroring similar abilities in athleticism, health maintenance and sexual vitality. More importantly, by describing the Flynn effect as a universal, continuous, incremental function of time (not of cohorts), the Griswold interpretation opens the door to a straightforward, environmental explanation for the Flynn effect. Any consistently changing phenomenon that is essentially present for all people in all places at all times becomes a viable candidate as a causal explanation for the Flynn effect, and such candidates are quite conceivable. I have described elsewhere that I believe it is the increasing amount of non-biological pattern, structure and form tangibly contained within the human environment that serves as the most likely driver for the Flynn effect, and although here is not the place to argue the merits of that explanation, it is permissible for me to note it is entirely consistent with the Griswold interpretation. I am unaware of any competing Flynn effect explanation that is entirely consistent with the Agbayani interpretation.
Agbayani, K.A. & Hiscock, M. (2013). Age-related change in Wechsler IQ norms after adjustment for the Flynn effect: Estimates from three computational models. Journal of Clinical and Experimental Neuropsychology, 35(6), 642-654.
Agbayani, K. A. (2011). Patterns of age-related IQ changes from the WAIS to WAIS-III after adjusting for the Flynn effect. Retrieved online from http://repositories.tdl.org/uh-ir/handle/10657/236.
Dickinson, M. D. & Hiscock, M. (2010). Age-related IQ decline is reduced markedly after adjustment for the Flynn effect. Journal of Clinical and Experimental Neuropsychology, 32(8), 865-870.
