It's been a few days, so I thought I would check in.
There are multiple other models for how/why substance abuse develops.
1) In regards to these articles
a) The authors redefine "self medication" however they want. This "hypothesis" is a specific thing.
b) The surveyors are asking people who have a substance use disorder to begin with, if their use is self medication. As you mentioned, there is a high degree of social bias which is for some reason mitigated by the self medication explanation.
c) First they propose self medication is considered a separate variable. Some people self medicate, some don't. Some do it before onset of psychiatric symptoms (which who the hell knows how one medicates against something before it starts). If "self medicating" behaviors are not directly attributable to a psychiatric disorder, then these behaviors are not self medicating.
d) lumping "drugs" into a homogeneous group completely redefines self medication. Again self medication is the use of a substance which has a property in the opposite direction of the affective disorder. Methamphetamine is not the same as benzodiazepines.
e) I didn't see where they addressed how the affective disorder starts and then the substance use starts.
f) The self medication hypothesis specifically states that once the underlying affective disorder is treated, the substance abuse will go away. This is clearly contradicted by this evidence.
g) Self medicating against a specific phobia, in many if not most instances, contradicts the idea of a specific phobia. And developing alcohol dependence from a specific phobia is evidence for alternative models, not the self medication hypothesis.
2) If substance abuse is self medication, we would expect the affective disorder to start first. If we look at the same data, we can see that based upon mean age of onset alcohol
dependence begins before the other affective disorder aside from social phobia and specific phobia. This would contraindicate the self medication hypothesis and support the substance induced anxiety model (e.g., Cohen, 1995; Katerndahl, 1999; Breese, 2005; etc). This data is similar to those found in the original Khantzian validation study.
3) Goodwin, 2004 shows that when extraneous factors are accounted for, anxiety is not a predictor for alcohol abuse.
4) IIRC, it was either the Grant or Helzer study from the 1980s using the national survey that contradicted the self medication hypothesis which led to Khantzian redefining his hypothesis. Because hypothesis can mean whatever now.