Time and punctuality
Posted: Fri Oct 13, 2017 5:46 amHi I had a previous thread that helped a lot and now I was hoping that I could find help here again.
We still plan an epxeriment to estimate demand elasticities for a certain transport mode. We use three alternatives (one opt out) 3 attributes and 24 choices in two blocks.
We identified price (6 levels), travel time (3 levels), and punctuality (3 levels) as the important attributes.
But now I am struggling with the best way to take account of punctuality.
We were thinking to include it as a percentage of punctuality as in 10 % of transports are punctual or as average expected delay in time.
The latter might not work well as respondents would tend to just add the expected delay on top of travel time which is not what we would expect them to do in real life, as delays are not only about the average delay but also how frequent and how extreme they are. We also thought to include the severity of the delay as an ordinal variable either as on time/ late / very late or never late / frequently late / very frequently late. But this would also not really account of the entire impact of punctuality.
I then though it would be best to define travel time as expected travel time including average delay and use the frequency of delays as a third attribute.
For the estimation we could then split travel time in on time travel time and delay depending on the third attribute. Respondents would always the same expected travel time (e.g. 10h), but depending on the level of punctuality 10h would be based on 10h pure travel time or 9h travel time + 1 h delay, or added later in the data set. Using an interaction affect I was hoping to estimate different effects of travel time on utility for each level of punctuality. But this would make the travel time variable and the delay variable correlated, which should be avoided.
I guess my question is: what is the best way to include travel time and punctuality while keeping a 3 attribute design and taking account of the different aspects of punctuality.
I am happy to elaborate more on the problem if my description was not clear enough.
We still plan an epxeriment to estimate demand elasticities for a certain transport mode. We use three alternatives (one opt out) 3 attributes and 24 choices in two blocks.
We identified price (6 levels), travel time (3 levels), and punctuality (3 levels) as the important attributes.
But now I am struggling with the best way to take account of punctuality.
We were thinking to include it as a percentage of punctuality as in 10 % of transports are punctual or as average expected delay in time.
The latter might not work well as respondents would tend to just add the expected delay on top of travel time which is not what we would expect them to do in real life, as delays are not only about the average delay but also how frequent and how extreme they are. We also thought to include the severity of the delay as an ordinal variable either as on time/ late / very late or never late / frequently late / very frequently late. But this would also not really account of the entire impact of punctuality.
I then though it would be best to define travel time as expected travel time including average delay and use the frequency of delays as a third attribute.
For the estimation we could then split travel time in on time travel time and delay depending on the third attribute. Respondents would always the same expected travel time (e.g. 10h), but depending on the level of punctuality 10h would be based on 10h pure travel time or 9h travel time + 1 h delay, or added later in the data set. Using an interaction affect I was hoping to estimate different effects of travel time on utility for each level of punctuality. But this would make the travel time variable and the delay variable correlated, which should be avoided.
I guess my question is: what is the best way to include travel time and punctuality while keeping a 3 attribute design and taking account of the different aspects of punctuality.
I am happy to elaborate more on the problem if my description was not clear enough.