Seminar über aktuelle Fragen zur Dynamik komplexer Fluide: Smog Alerts in Stuttgart: Impact Analysis of Soft Transport Policy using State Space Models
Seminar über aktuelle Fragen zur Dynamik komplexer Fluide
- Date: May 18, 2018
- Time: 10:15 AM - 11:15 AM (Local Time Germany)
- Speaker: Anne Berner
- NGM, MPI-DS
- Location: Max-Planck-Institut für Dynamik und Selbstorganisation (MPIDS)
- Room: SR 0.77
- Host: MPIDS/DCF
- Contact: stephan.herminghaus@ds.mpg.de
Many large cities worldwide are nowadays facing the problem of smog and particular matter, that goes along with urbanization and industrialization. While some cities try to tackle the smog issue with car bans, the city of Stuttgart implemented an air pollution alert in 2016. This soft policy measure builds on voluntary incentives to promote more sustainable means of transport as well as to create awareness of the issue. Even though it is widely discussed in public whether the alert can convince people to change their traffic behavior, a detailed quantification of the effect on traffic volume is missing.
We analyzed the causal impact of the policy intervention on traffic volume using a Bayesian Dynamic Linear Model. We applied a learning forecast algorithm, the Kalman filter, to predict and continuously update counterfactual scenarios indicating the traffic flow that would have occurred in the absence of the smog alert. The causal impact of the smog alert was deduced by comparing the observed trajectory to the estimated counterfactual. In order to control for some regional or global intervening factors, traffic data from Karlsruhe was used as a control time series.
We found that traffic volume is only sparsely mitigated by the smog alert. Further we observed a time-dependent habituation effect, whereby the negative impact of the initial alert of each “alert season”, starting in fall, exceeds the impact of all subsequent alert event. This is reminiscent of aspirational behavior following new year resolutions.
We analyzed the causal impact of the policy intervention on traffic volume using a Bayesian Dynamic Linear Model. We applied a learning forecast algorithm, the Kalman filter, to predict and continuously update counterfactual scenarios indicating the traffic flow that would have occurred in the absence of the smog alert. The causal impact of the smog alert was deduced by comparing the observed trajectory to the estimated counterfactual. In order to control for some regional or global intervening factors, traffic data from Karlsruhe was used as a control time series.
We found that traffic volume is only sparsely mitigated by the smog alert. Further we observed a time-dependent habituation effect, whereby the negative impact of the initial alert of each “alert season”, starting in fall, exceeds the impact of all subsequent alert event. This is reminiscent of aspirational behavior following new year resolutions.