Hey, I took way too long to write a bachelors thesis on an empirical study I did with Eclipse/C4DT and OpenClonk development, measuring impact of code completion and solving time for some small scenario development tasks. There were 10 test subjects who actually finished the coding tasks (such significance) and I kind of messed up writing the actual paper and results were not clear enough but whatevs.

Design was:
Three different modes: C4Script with ominous homegrown type inference (c4dt default, i), C4Script with type annotations (such as var Clonk clonk; func int blablu() {}, s), both with instant code completion and C4Script without type inference and no code completion (n).

Two groups with different mode setups where participants had to solve variations of the same four tasks:
Group 1: s, n, i, i
Group 2: i, s, n, s

Results: Tendency towards code completion helping yay?
Tasks: https://bitbucket.org/mortimer/bachelors-study-excercises/src/tip, those actually used were Waypoints, Dialog, Vegetation, ExplosiveRain, <scen>.ocd/Task.ocd/Script.c was where people had to type in stuff (empty func Task() {})

pdf: Link