Treatment-Effect Identification Without Parallel paths An illustration in the case of Objective 1- Hainaut/Belgium, 1994-2006

Imagine a region that has a lower income per head than the rest of the territory, and becomes eligible for a generous EU-funded transfer programme (the treatment). The evaluation of the effectiveness of such a policy can rest on a difference-in-differences analysis (DiD); which in essence consists of comparing the income-level handicap before and after the treatment. Imagine that DiD shows that the handicap has not diminished, or even that it has risen. Most observers would conclude to the inefficiency of the policy. The point we raise in this paper is that second thoughts are needed, because DiD rests heavily on the validity of a key assumption: parallel paths in the absence of treatment. Without EU money the outcome difference between the treated and the controls should be time-invariant; so that any statistically significant change of that difference can be ascribed to the treatment. Parallel paths seems a priori unrealistic in the context of old industrial regions, as one of the reasons they become eligible for treatment is that their income-level handicap is on the rise. Also, from a methodological point of view, when more than one pre-treatment period is available in the data, the parallel-paths assumption can easily be abandoned in favour of more flexible assumptions as to the relative dynamics of treated vs. control entities. This paper illustrates the relevance this approach in the case of Objective 1-Hainaut; an EU-funded transfer policy implemented between 1994 and 2006 in Hainaut, the most economically-deprived province of Belgium; a former bastion of the country’s industrial revolution that has endured decades of decline.