Skip to main content


Table 4 Impact of results-based financing on location and assistance of deliveries: Results from the segmented linear regression models

From: Boosting facility deliveries with results-based financing: a mixed-methods evaluation of the government midwifery incentive scheme in Cambodia

Dependent variable Facility deliveries in districts with no other major financing scheme Facility deliveries in districts with one or more other major financing scheme Facility deliveries in all districts All deliveries by trained health personnel in all districts
Number of health districts 19 58 77 77
Model (1) (2) (3) (4)
Constant 610.509*** 3,710.119*** 4,315.636*** 10,366.858***
  (413.080–807.938) (3,132.182–4,288.055) (3,515.163–5,116.110) (9,364.438–11,369.278)
Time(month) 36.235*** 93.407*** 132.135*** 125.913***
  (20.380–52.090) (47.549–139.264) (70.328–193.942) (53.338–198.488)
GMIS Intervention 489.685*** 912.637*** 1,330.918*** 1,260.934**
  (219.206–760.163) (306.590–1,518.685) (488.316–2,173.519) (181.964–2,339.903)
GMIS postslope −1.307 83.973** 80.773* 54.600
  (−19.326–16.712) (16.035–151.911) (−5.097–166.644) (−45.848–155.049)
Observations 72 72 72 72
R-squared 0.792 0.732 0.740 0.624
Durbin Watson original 0.985 0.688 0.711 0.792
Durbin Watson transformed 2.043 1.898 1.909 1.940
  1. all regressions are using a Prais–Winsten estimator that corrects for data auto-correlation; *** p < 0.01, ** p < 0.05, * p < 0.1; confidence intervals (CI) in parentheses. Time variable is a sequence starting at 1 for the first month of the dataset (January 2006) to 72 for the last month (Dec 2011), its coefficient provides the secular trend of deliveries. GMIS Intervention and GMIS postslope are the level and trend variables for an intervention starting in October 2007: their coefficients represent respectively the change in level and the change in trend of deliveries after the introduction of GMIS. Other major health financing schemes include contracting and other performance-based financing, health equity funds, vouchers and community-based health insurance. R-Squared gives information about the goodness of fit of the model, the closer to 1, the better the data fit the model. Durbin-Watson (DW) statistic tests the presence of first-order auto-correlation. The presence of first auto-correlation violates the ordinary least squares (OLS) assumption that the error terms are uncorrelated, meaning that the standard-errors and p-values are biased with the OLS estimator. DW ‘original’ tests the presence of first-order auto-correlation with the OLS estimator, while DW ‘transformed’ tests it with the Prais-Winsten estimator. A value around 2 indicates no sign of auto-correlation. P-values and CI are based on a standard variance estimator.