 Let's summarize what we have discussed about measurement. The important concepts in measurement are reliability and measurement validity. Reliability simply means that if you measure the same subject or same company, again you will get the same result, so it's about consistency of measurement and the lack of random noise. Validity on the other hand is much more complicated, it is about whether the indicators measure what they are supposed to be measuring. One way to think about that is that the measured attribute must exist and it must causally produce variation in the observed score, so that's one definition of validity. That is of course more difficult to demonstrate. Reliability can be demonstrated empirically because it's basically a statistical concept, lack of random noise. You can either repeat the same measurement over and over and check how much does correlate. That will be evidence of reliability if you can argue that the measurements are actually independent and that the trait that you're measuring has not changed. It is possible to do if there's a time delay, but in that case you have to set the time delay to be sufficiently long so that the person doesn't remember that they were measured before. If you have a bathroom scale and you measure a person on a bathroom scale, then when the person steps off the scale, the scale resets in a couple of seconds and then the person can be measured again. If you are talking about psychological measurement, asking people questions, then you may need to give the person a few days or a few weeks time to reset and then you ask the question again. So that's the test-tree test approach. Another way of doing reliability is do parallel tests, in which case you use different measures that are sufficiently different so that you can argue safely that they are actually distinct instead of just repeating the same measurement over and over and that they measure the same trait. Then you can check the correlation between those two indicators and if they are unidimensional, then the correlation is an indication of reliability. Multiple measurements can be also used to improve reliability. So the idea is that if you have unreliable measurement like you measure on the weight of a wiggling tile using a bathroom scale, then taking multiple measurements and taking the mean is more reliable than any individual measurement. Demonstrating validity is much more difficult problem and it cannot be done directly using statistical technique. We can demonstrate validity in two different ways. We can only demonstrate that our observations are consistent with our measurement theory. So the idea of construct validity was that if a measure for innovativeness is valid measure for innovativeness, then it should be positively associated with any measure of any possible consequence of innovativeness. So the idea of construct validity was whether the theoretical correlations and empirical correlations agree. But ultimately measurement is about making a causal claim. The measurement is valid if the variation in the construct causes variation in the indicators. That is not a meteorological question. Instead it is a substantive question that cannot be fully outsourced to meteorology. So ultimately you have to be able to claim in which way for example why would a person are from a more innovative company answer positively to a question about innovativeness than a person who works in a less innovative company.