Internal Validity (IV): The extent to which we can say that the effect is actually caused by the manipulation. The key point is to take good control of manipulation implementation & Extraneous Variable. Field experiment is inevitably bad for IV but good for External Validity (EV)–the extent to which the effect found can be replicated or generalized to the real world settings.
The internal validity is usually a mess, especially in the education research field, partially because of the extreme complexity regarding human brains and minds; partially because of a variety of different context and goals education could have. The problem not only lies in the control of independent or dependent part, but most seriously in the extraneous variables that are literally uncontrollable or even unnoticeable in the education field. Worse yet, to overcome this intrinsic complexity issue, the only way is to go big and make use of the large sample size to balance the uncontrollable bias. As a result, people go field experiments for more samples at lower cost, and in turn further worsen the internal validity.
In my research, we utilize the power of information technology. By moving the whole class online, we are able to track much more detailed information, especially the learning behavior information–which is of much higher quality compared to academic test or survey when it comes to scientific research. In my humble opinion, this is actually the beginning of a phenomenal revolution that opens a brand new door for education researchers to get to the truth.
Types of Validity
Validity has two different topics: Experiment Validity (EV) & Measurement Validity (MV).
To achieve EV, first secure internal validity, then external validity, and finally conclusion validity. Conclusion validity is supported by quantitative, qualitative, and all other validities.
MV involves face, construct, content, criterion validity. Construct validity includes Convergent and discriminant validity.
Ref: https://en.wikipedia.org/wiki/Validity_(statistics)