Finance and Economics Discussion Series: Accessible versions of figures for 2024-096

A Market Interpretation of Treatment Effects

Accessible version of figures


Figure 1: 1a. Industry-level effects of productivity treatments
1b. Firm-revenue effects of productivity treatments
Note: DiD is the sum of the two red revenue amounts. The two green revenue amounts represent market spillovers.

Figure 1 shows how the difference-in-differences estimate of a treatment effect can have the opposite sign from the effect of treating the entire market. Figure 1a shows how treating more suppliers with a productivity treatment shifts out market supply curves (not shown), resulting in higher quantities E’ and E’’ and lower prices P’ and P’’. Figure 1b shows revenue rectangles that treated and untreated suppliers receive under different assumptions on the share of suppliers that receive treatment. Before treatment, for example, each supplier receives revenue R₁ + R₃ + R₅. When demand is inelastic, this is greater than the revenue each supplier would receive if all suppliers received treatment, R₅ + R₆. Nevertheless, the difference-in-differences estimator comparing revenues of treated suppliers to untreated suppliers always leads to a positive estimate, R₄ + R₆.

Return to text.


Figure 2: Treatments and Market Spillovers
For the purpose of Figure 2 the controls are untreated (dk=0) but in the same market. Observations reflect a treatment of dt=1 directly affecting the treatment group

Figure 2 illustrates our model setup, which allows for both treated T and control K to be affected by treatment and other influences θ.

Return to text.


Figure 3: Geometry of Market Spillovers

Figure 3 shows an example in which the scale effect exceeds the difference-in-differences estimate.

Return to text.


Figure 4: Market does not easily deviate from \(45^{\circ}\)

Figure 4 shows how the degree to which the scale effect can exceed the difference-in-differences estimate can be extreme.

Return to text.


Figure 5: Scale effect = 0 <<DiD

Figure 5 shows that the difference-in-differences estimate can be large when the scale effect is 0.

Return to text.