Aspiration formation and satisficing in search with(out) competition

Abstract

We experimentally explore individual and competitive search, and we test whether generally accepted principles of bounded rationality adequately explain observed search behavior. Subjects can, at a cost, employ screening and selection methods not only facilitating search but also directly revealing their aspirations. Most subjects follow the single threshold heuristic after extensive experimentation. Surprisingly, aspiration levels are set below the maximum value of all previously inspected alternatives. In competitive search, subjects tend to experiment less before engaging in satisficing and generally state lower aspirations. Finally, systematic satisficing seems to pay off.

Keywords:

• sequential search
• secretary problem
• optimal stopping
• bounded rationality

Acknowledgements

The authors gratefully acknowledge the helpful advice of two anonymous referees and their editor Simon Kemp.

Notes

1. See Bolle (1979) and Zwick et al. (2003), to mention just a few. One simply assumes that after exploring a number of alternatives participants use them to form an aspiration level. Several heuristics, based on aspiration building, were tested for validity (see, for example, Dudey & Todd, 2001; Todd, Rieskamp, & Gigerenzer, 2008).

2. See Zwick et al. (2003), Kogut (1990), and Rapport and Tversky (1970).

3. This and alternative payoff functions are frequently encountered in generalized secretary problems and discussed in more detail by Bearden and Murphy (2004).

4. This differs from the theoretical analysis of Alpern and Gal (2009) who assume that only one secretary is hired by a selection committee whose preferences are partly conflicting.

5. Gilbert and Mosteller (1966), Seale and Rapoport (1997, 2000), to mention just a few.

6. For a comprehensive description of the multiple threshold rule, see Lindley (1961) and Chow et al. (1964), who extensively discuss the class of MTR search heuristics.

7. The simulation was conducted using the open source statistics package R.

8. The width of each box plot is determined by the log value of the relative frequency of observations it is based on.

9. The share of participants, who actually achieved their previously encountered maximum, is 47.3% in IK, 51.2% in IU, and lowest with only 35.9% in CK. Of course, when using the A-routine, what one encounters is only the last candidate and one never encounters the candidates after stopping search. The globally best candidate is hired in treatment IK with 23.4%, in IU with 16.0%, and in CK with 16.0%.