Shifting education standards could be to blame for the drop. It is well-known that each additional year of school -- including higher education -- adds a few points to someone's IQ score. (It maybe adds less to people at the upper ranges.) And perhaps changes in education can sometimes reduce the score -- e.g. less classroom meeting time; less rigorous curriculum; over use of calculators and search engines.
Education was associated with an average increase of 2.7 points per year on the Armed Forces Qualifying Test, taken in the late teens or early twenties. This number was broadly concurrent with a later study by Falch and Sandgren Massih (2011) that analyzed data from the Malmö Longitudinal Study (initial n = 1,547). They found that, controlling for ability at age 10, education improved IQ, measured on a test designed to be similar to the early measure, by 2.9 to 3.5 points per year by age 20.
Using two alternative econometric analyses (difference-in-difference and instrumental variables) on a sample of over 100,000 individuals, Brinch and Galloway (2012) estimated the benefit of 1 year of schooling at 3.7 IQ points on average.
As to the genetic influence on differences in IQ, a few genes have been identified; but the studies are mostly correlational -- the causal effects aren't completely understood.
Twin studies show that even identical twins can differ by 20 IQ points or more, perhaps due to environmental effects or due to "random wiring" at birth (genes don't give an exact blueprint for what all the neurons are supposed to do, and exactly how they should fit together.)
As I recall, in one study the standard deviation in IQ among identical twins is about half that for random pairs of people from the population. This has been used to argue that epigenetics has only a small effect. However, it was recently discovered -- much to the surprise of scientists -- that the epigenome of identical twins is identical, or nearly identical, at birth. This makes it difficult to separate genetic effects from epigenetic ones:
The epigenetic variation at MEs is determined randomly and is influenced by many environmental factors, ranging from the nutritional breakdown of the mother’s diet to the season. Consequently, it was expected that levels of epigenetic similarities and differences for MEs would be similar for both identical and fraternal, or non-identical, twins.
What they found was something of a shock.
Their research, published in Genome Biology, shows that monozygotic twins have identical epigenetics at MEs. “We found that the methylation patterns matched almost perfectly in identical twins, a degree of similarity that could not be explained by the twins sharing the same DNA,” says Waterland. “We call this phenomenon 'epigenetic supersimilarity.'”
One last thing worth pointing out, about genes for intelligence:
But other gene studies have shown that variants in one population can fail to predict what people are like in other populations. Different variants turn out to be important in different groups, and this may well be the case with intelligence.
“If you try to predict height using the genes we’ve identified in Europeans in Africans, you’d predict all Africans are five inches shorter than Europeans, which isn’t true,” Dr. Posthuma said.
This is the same problem faced by Machine Learning practitioners, where a model does very well on media taken from a specific distribution, but does very poorly if you apply it to data from a slightly different one.