Experiments in Agile Estimation was a workshop, facilitated by Mike Hill, Nils Haugen and Stein Grimstad, that explored planning poker as a group estimation technique. For the first experiment we had to estimate user stories on our own. It looks like I have a tendency to be optimistic. Perhaps I'm too courageous? Or just stupid? Don't answer that! The second experiment explored how effective planning poker was with respect to individual estimation.
The statistical evidence collected to date (as part of the research conducted by Nils and Stein) was not presented very clearly (take a look at this pdf). But they observe a tendency to overestimate when using planning poker. Nils asked an insightful question: Is it actually a tendency to finish under the estimate? Looking at the results from an analysis of the code written on the projects used in the research sample, work estimated with planning had, on average, twice as many deleted control statements (perhaps indicating an effort to do the simplest thing) and twice as many deleted out-of-class references (perhaps indicating an effort to reduce coupling). So, does planning poker positively affect the quality of implementation? I'm not convinced.
Planning poker is definitely fun, and lets face it, estimation is something we all hate doing. It gets everyone involved and talking face-to-face. And the group discussions draw knowledge from many sources, help to explore uncertainties and reveal more hidden detail before estimating. Planning poker incorporates participatory decision making and is effectively consensus-driven. Reaching an estimate as a group gives developers a sense of comfort as they collectively own the estimate and it also gives them more confidence in the estimate. The technique can feel cumbersome when you first start doing it. However, once people get used to it the actual overhead is very low. But be careful that discussions don't take too long.
One person at the session thought the use of cards was unnecessary. I disagree. Always use the cards. They make it clear that everyone is engaged and participating and it helps retain a structure to the proceedings. Without structure, group discussions can quickly become wasteful.
Originally, we used index cards with numbers written on them. Now we're using nice glossy cards that Conchango were handing out at the Agile Business Conference. We're using ideal pair days as estimation units (and not story points) so the only cards we use are 0, 1/2, 1, 2, 3 and ?. User stories that get a ? are split into smaller stories.
As the discussion tails off, the facilitator says "Ready" and asks "Do you have enough information to estimate this story?".
When everyone responds "yes", the facilitator says "Steady. Select your estimate." Each person then selects their card and holds it out in front but does not reveal the number to the rest of the team.
When everyone is holding a card out, the facilitator says "Go" and everyone reveals their estimate by turning their card around to face the team. (Simultaneous revealing of estimates might reduce an anchoring effect where people who reveal their estimates early influence estimates from others in the team.)
The facilitator asks the people with the lowest and highest estimates to explain their selection.
The facilitator says "Lets do another round". Go back to 4.
Loop through points 4 to 7 until all the estimates converge at a single number.
Previous Topic
.