You cannot really improve your IQ score by any significant margin by studying in general, unless you memorize the answers on a particular test. This doesn’t make the IQ tests any less useful or valid than SAT or MCAT.

Scores should be reproducible.

They are. Test-retest correlation on high quality IQ tests are well north of 0.9

I've yet to see an IQ test that didn't test skills you could improve by practicing.

If you practice (I.e. memorize the answers) for a specific test, e.g. Wechsler, you won’t get any gain on Stanford Binet, for example.

They also assume knowledge. For example, the IQ test I had when I was young included word comparisons. This tested vocabulary more than intelligence. Someone who didn't have a formal education or didn't know English well would score lower regardless of their intelligence.

Yes, because intelligent people tend to have more knowledge. People don’t attain vocabulary simply by being taught the words at school. Vocabulary is learned by experiencing contact with words, and intelligent people tend to seek these experiences more. But yes, comparing an Englishman and Korean on an English-language analogies section of WAIS makes no sense. Fortunately, there almost always is measurement invariance within same country/culture, meaning that the tests measure the same latent ability.

They're not infinitely adaptable, though. The further you go from whatever baseline you're working with, the harder it will be to make significant progress. What's more, people with higher base scores will tend to progress faster. These repeat effects also tend to fade over time (https://eric.ed.gov/?id=ED076675). Finally and most importantly, virtually all types of skill training are extraordinarily domain-specific. In the digit span example above, the individual who trained up to a digit span of 79 was tested on chains of alphabetic characters instead and went right back down to the standard 7 +/- 2.

What does all this mean for IQ tests? Well, the goal is to identify broad baselines, not specific trained skills. The best ones will test a broad range of specific cognitive skills (like the Stanford-Binet) or a specific g-loaded skill that an individual hasn't practiced (like Raven's Progressive Matrices). It's fine for them to assume knowledge if they are directed towards specific cultures (for example, English vocabulary is fair game in an IQ test aimed at teen/adult native English speakers), but tests like Raven's should be used for broader contexts. If someone really needs an accurate idea, comparing several tests is worthwhile: A determined individual can practice some specific skills on one test, but those skills won't transfer.

IQ isn't perfect, but it's a useful proxy for intelligence that correlates meaningfully with quite a bit. Specific IQ tests can be trained for in limited ways, but such training only reflects narrow skills and so can easily be avoided by testing a variety of skills in thoughtful ways.

[0] - https://en.wikipedia.org/wiki/G_factor_(psychometrics)#Pract...