A Controlled Experiment on the Process Used by Developers During Internet-Scale Code Search - Finding Source Code on the Web for Remix and Reuse - page 65

Fig. 4.4 Effect of intention on P@10

a b

0.70 0.70

Block
Subsystem

0.53

0.53

0.35 0.35

0.18 0.18

0 0

Google Koders

Overall P@10 for Search Engine

Krugle GCS SourceForge Google Koders Krugle GCS SourceForge

P@10 for Search Engine x Size

Fig. 4.5 Effect of search engines and interaction effect search engine and size

= . The highest rate occurred when using Google, whereas the low-

est rate occurred with SourceForge, as shown in Fig. 4.6 a. An interaction effect was
found between intention of the search and the search engine

6 . 44 , p < 0 . 01 )

( F ( 4 , 100 )= 2 . 55 , p <

0 . On SourceForge, the clickthrough rate was higher when searching for code

to reuse as-is than for reference examples; the opposite was true for all the other
search engines.

. 05 )

4.4.4 Time Spent

The search session as a whole was characterized by time spent. We found that de-
velopers spent an average of 6 min on any search engine for searching. An ANOVA
revealed three significant effects (Fig. 4.7 ).
There was a main effect from the intention of search ( F ( 1 , 100 )= 4 . 66 , p <

0 . More time was spent looking for reference examples (417 s) than code to

reuse as-is (334 s). The size of the search target also affected the duration of the

. 05 )