Using CausalImpact for severalt units treated at different times: health project in care centers

[f.bede...@gmail.com]

Dear colleagues,

The short version of my question is: Can I use this package to evaluate the impact of one treatment on several units (x1, x2,...,x61) begining their treatment at different point in time, using a large counterfactual (y1, y2,...,y400+).

The longer version is: We are evaluating an development program aimed at reducing maternal and child mortality in a subsaharan country. It consists in offering a voluntary insurance scheme to pregnant women. A fixed premium (~20 USD) entitles women to an obsteric package including 4 pre-natal consultations, prophylactic treatments, blood and urine tests, ulta-sound, care at delivery and any complication (cesarean, ambulance transfer, resucitation unit) and post-natal care. Piloted since 2002 in the capital city hospitals, this scheme is being progressively extended to other health care facilities since 2007. As of end 2013, we had 92 health care facilities covered, each one with high take-up (>90% of deliveries are covered by the insurance in the health care centers offering it).
From the national health information system data, we have the monthly assisted birth reported in all the 500 health care centers in the country from 2009 to 2013, including 61 health care centers that adopted the insurance scheme in this period. I wonder if I can use this package to evaluate to which extent the insurance scheme increases the access to assistance at birth in the health care centers offering it.

Thanks in advance for your feedbacks,
Best regards,
FB

[Kay Brodersen]

Hi Florent,

One option would be to align all time series (treatment and control) at the time of treatment. You could then sum your response time series and infer the causal effect on the sum, corresponding to a fixed-effects analysis. Another option would be to run 61 separate analyses, summarise each analysis in terms of overall impact (a single number) and then consider the distribution of this statistic across all 61 units, corresponding to a summary-statistics approximation to a mixed-effects analysis.

With best wishes,

Kay

--
You received this message because you are subscribed to the Google Groups "CausalImpact" group.
To unsubscribe from this group and stop receiving emails from it, send an email to causalimpact...@googlegroups.com.
To post to this group, send email to causal...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/causalimpact/678f29b6-a154-4667-ad74-1eb8ed2cfea9%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.