I keep hearing that intelligence is hard to measure and that IQ tests are imperfect tools. But I also know that IQ scores measured in childhood correlate strongly with scores measured decades later and predict a huge range of life outcomes. If the measurement were truly difficult and unreliable you would not expect that kind of stability and predictive validity. So which is it, is intelligence hard to measure or not?
The apparent contradiction dissolves once you separate two different meanings of difficult. Difficult to measure perfectly is not the same as difficult to measure usefully. IQ tests are imperfect in the sense that observed scores contain measurement error and do not capture the full latent construct with complete precision. But the signal is strong enough that scores remain remarkably stable across decades and predict outcomes better than almost any other single measure in social science. The recent Herasight polygenic score research actually quantified this directly, showing that correcting for measurement error in cognitive assessments substantially increases the estimated correlation with underlying genetic cognitive ability. The true construct is even more stable and heritable than raw scores suggest 
A speedometer with a slight calibration error is still useful for driving 
Imperfect measurement and useless measurement are completely different things. IQ tests are imperfect the way every psychological measure is imperfect, but the signal they capture is real, stable, and remarkably predictive 