It's one thing to compare different call scheduling algorithms. You can compare two algorithms and measure the performance using whatever metrics you want to compare (efficiency, response rate, survey outcome variables).
But what about comparing estimated contact propensities? There is an assumption often employed that these calls are randomly placed. This assumption allows us to predict what would happen under a diverse set of strategies -- e.g. placing calls at different times.
Still, this had me wondering what a really randomized experiment would look like. The experiment would be best randomized sequentially as this can result in more efficient allocation. We'd then want to randomize each "important" aspect of the next treatment. This is where it gets messy. Here are two of these features:
1. Timing. The question is, how to define this. We can define it using "call windows." But even the creation of these windows requires assumptions... and tradeoffs. The key assumption about a window is that contact probabilities within any window are homogenous. We can make very wide windows (i.e. windows with big chunks of time). These windows will have more data in each window. But the assumptions that contact probabilities are homogenous within any window seems plausible. If we make narrow windows, then the homogeneity assumption is more plausible. But we have less data in each window. Imagine estimating contact probabilities across 24*7=168 windows, one for each hour of the week!
2. Lag. How much time between each call? Most call centers and field operations don't do a great job controlling this dimension. Some cases may have huge lags. It may be hard to explain way. For some reason, they fall to the "bottom of the pile" and don't get called very frequently. Again, what are the appropriate lags?
So, small studies have very little ability to estimate a large number of windows and/or a large number of lags. Let alone constrain production to true randomization of these features. Still, for the sake of methods research, it might be fun to try this.
But what about comparing estimated contact propensities? There is an assumption often employed that these calls are randomly placed. This assumption allows us to predict what would happen under a diverse set of strategies -- e.g. placing calls at different times.
Still, this had me wondering what a really randomized experiment would look like. The experiment would be best randomized sequentially as this can result in more efficient allocation. We'd then want to randomize each "important" aspect of the next treatment. This is where it gets messy. Here are two of these features:
1. Timing. The question is, how to define this. We can define it using "call windows." But even the creation of these windows requires assumptions... and tradeoffs. The key assumption about a window is that contact probabilities within any window are homogenous. We can make very wide windows (i.e. windows with big chunks of time). These windows will have more data in each window. But the assumptions that contact probabilities are homogenous within any window seems plausible. If we make narrow windows, then the homogeneity assumption is more plausible. But we have less data in each window. Imagine estimating contact probabilities across 24*7=168 windows, one for each hour of the week!
2. Lag. How much time between each call? Most call centers and field operations don't do a great job controlling this dimension. Some cases may have huge lags. It may be hard to explain way. For some reason, they fall to the "bottom of the pile" and don't get called very frequently. Again, what are the appropriate lags?
So, small studies have very little ability to estimate a large number of windows and/or a large number of lags. Let alone constrain production to true randomization of these features. Still, for the sake of methods research, it might be fun to try this.
Comments
Post a Comment