Sequence effects in the estimation of software development effort
This is a post based on the publication “Sequence effects in the estimation of software development effort” by MagneJørgensen and TorleifHalkjelsvik.
Please read the full publication for further details. Here are the key findings where justinagile has added images to help visualise.
Anchoring:
People’s judgments may be affected by judgments made immediately before (Sharif and Oppenheimer, 2016), and that prior judgment may act as an “anchor” making the prior and subsequent judgements more similar (Mochon and Frederick, 2013).
Under-estimation:
Knowing more about the presence and nature of sequence effects in software development effort estimation may be useful to guide the estimation work. In particular, it may be useful to guide the estimation sequences to avoid biases towards strong under-estimation, given the tendency towards too low effort estimates in software development (Budzier and Flyvbjerg, 2013).
Two types of sequence effects:
The first effect, when a judgment is biased towards becoming more similar to the previous one, is called an assimilation effect, while the second effect, when a judgment conducted immediately after a previous judgment is biased towards becoming more different than the previous one is called a contrast effect (Wedell et al., 2007; Mussweiler, 2001).
The domination of the contrast effect:
Increased effort estimates when estimating tasks of similar size in a sequence.
The domination of the assimilation effect:
Estimates pulled towards that of the previous task when estimating tasks with more different sizes.
Underestimation of effort:
Estimate the effort of a larger task immediately after a smaller one.
This sequence is, because of the assimilation effect, likely to increase the risk of underestimating the effort of the larger task.
Estimating a larger task after a smaller one led to a mean decrease in the estimate of 24%.
Underestimating of effort:
Estimation of a smaller task just after a larger one.
Because underestimation is known to decrease productivity (Nan and Harter, 2009).
Estimating a smaller task after a larger one led to a mean increase of 25%.