Does Productivity Always Decrease With Higher Staff? A Project Management Paradox Explained

Does Productivity lwys Decrese With Higher Stff? Project Mngement Prdox Explined

In softwre , lrger softwre pplictions tke more effort to develop thn smller ones. The bigger the softwre size, the more effort. Tht see resonble, nd is wht we would expect. Wht isn’t obvious is tht the reltionships between the core metrics re ll exponentil. For exmple, the reltionship between size nd effort is logrithmic.

This reltionship cuses some surprises. For exmple, softwre productivity (size produced per unit of effort) rises s softwre size rises. QSM’s dt (>10,000 softwre projects) definitely shows n upwrd trend in productivity s ppliction size increses. This is true whether we use mesures like QSM’s PI (Productivity Index) or rtio bsed productivity mesures (e.g., SLOC or Function Point per person month of effort).

I took nother look t productivity dt.The follow-up question I nswe ws, “Does productivity (mesu s SLOCPM) lwys increse with system size, or could the size-productivity reltionship ctully behve differently in certin regions of the size spectrum?” To nswer this question I used stndrdized residuls to evlute the sizeproductivity regression trend.

Simply put, residuls mesure the difference between picted vlues (the vlue of the regression trend t prticulr size) nd ctul metric vlues. If the regression line provides poor “fit” in certin size regimes, the residul vlues will reflect the gp between the vlues picted by the trend nd ctul productivity vlues for tht size regime.

s cn be seen in the figure bove, the residuls form n lmost perfect norml distribon.  This implies tht there ws no unexplined skew in the dt.

Effort, productivity nd stff size ll tend to be higher on lrger softwre size projects. This cn be seen in the following figure, which uses over 4,000 projects completed between 2001 nd 2011.

Previous reserch (e.g., see rmel in Resources) hs shown tht lrge tem sizes (higher stff) results in lower productivity. 

So, lrge projects hve higher productivity. nd lrge projects hve higher stff. But higher stff results in lower productivity. How cn this be?

To understnd the underlying reltionships, we need wy to visully exmine three vribles t once: size, productivity, nd stff.

Cluste boxplots provide view of the trends. box in boxplot represents the interqurtile rnge of the dt. The bottom of the box is t the ft qurtile (25th percentile). The drk line inside the box is the medin (50th percentile). The top of the box is the third qurtile (75th percentile). The “whiskers” extending out from the box represent the rnge of the vlues. Individul outliers (if ny) show up s circles, nd extreme vlues re sterisks.

To crete the following plot, the projects were ft divided into qurtiles for size nd lso in qurtiles for pek stff. Qurtile 1 hs the smllest 25% of projects nd qurtile 4 hs the lrgest 25%. Productivity on the verticl xis is expressed on log scle to further improve the redbility.

In the bove plot, productivity decreses s pek stff increses within ech qurtile of size. To see this, pick ny of the size qurtiles, nd compre the position of the 4 djcent boxes. In the next grph, this hs been done with n ovl drwn round the second qurtile of size. Productivity drops s stff increses, for given size.

Next, we cn see tht within ech qurtile of pek stff, productivity increses s size increses. Pick ny color of box, nd compre the position of the 4 boxes with the sme color (one from ech qurtile of size). For given stff size, productivity is higher on lrger projects. In the following grph, the lrgest stff sizes re identified with rrows.

Simple productivity is higher on lrger softwre projects. Smller tem sizes tend to hve higher productivity. With this dt set, we’ve shown tht these two sttements re not mutully exclusive. Lrger te become more productive s project size increses, but productivity increses even further s tem size decreses.

For dditionl informtion on this topic, plese tke look t the QSM Benchmark Table nd the other resources ed below.

bout the uthor: Pul Below hs over 30 yers of experience in technology mesurement, sttisticl , estimting, Six Sigm, nd dt mining. s Principl Consultnt with QSM, he provides clients with sttisticl of opertionl performnce, process improvement, nd pictbility. He is Six Sigm Blck Belt, nd hs one US Ptent.

Does Productivity lwys Decrese With Higher Stff? Project Mngement Prdox Explined