Knocking out the Supply and Sorting in Decentralized Job Markets Guillaume Haeringer Universitat Aut` onoma de Barcelona Vincent Iehl´ e Universit´ e Paris Dauphine N´ uria Rodriguez-Planas Universitat Aut` onoma de Barcelona Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 1/31
Introduction Many-to-one markets using Deferred Acceptance with the “many side” proposing: students and schools (e.g., NYC, Boston); students and universities (Spain); job candidates and academic departments (France!!!!!). Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 2/31
Introduction Strategic behavior: Proposing side: not an issue, incentive to be truthful (really?) Receiving side: the game is manipulable. Some theoretical characterizations of optimal play has been proposed (Roth and Rothblum (1999), Ehlers (2004), Coles (2009)). In practice? Little is known. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 3/31
Introduction Questions: How do actors on the “receiving side” behave? How does this behavior affect market outcomes? Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 4/31
Introduction This paper: Describes the French Academic job market (junior and senior levels), link it to the Deferred Acceptance game. Documents on the behavior (and the market outcome) of the job market for junior mathematicians. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 5/31
The French job market Since 1984 the French job market is a centralized market. A year before the market: 1. Departments negociate with the universities about the openings 2. Universities negociate with the Ministry of higher education 3. Ministry makes the fi nal decision. ⇒ Departments have little control over the number of openings. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 6/31
The French job market The job market is organized into “sections”, one per fi eld (or sub fi eld). There are 74 different sections (122 if including medical sub-sub-sections): 5 = economics; 32 = organic, mineral and industrial chemistry 25 = (pure) mathematics, 26 = applied mathematics. Most positions are advertised for 1 section, but some positions are open to different sections. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 7/31
The French job market December-January: Candidates (with a PhD degree) apply to a national committee for the “right” to enter the market (valid for 4 years). Candidate OK for section X can apply to section Y (but usually not considered). February: job openings are anounced. Candidates send their package to the departments March-April: Departments announce the list of candidates to be interviewed. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 8/31
The French job market May: Interviews take place (3 weeks period). Departments rank the candidates (up to 5 per position). June: Candidates submit their “preferences”. End of June-early July: fi nal assignment published. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 9/31
The French job market The “matching algorithm” used by the French Ministry combines three algorithms: 1. Top-top match : Match a candidate and a department if they are each other’s fi rst choice. Update candidates’ submitted preferences and Departments’ rankings, and repeat. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 10/31
The French job market If “top-top” does not produce a complete matching: 2. “Clean the lists” : Delete the vows that will never be expressed. Example: Candidate i ranked k -th at university X , and X is i ’s fi rst choice. Delete the candidates ranked after i at X in X ’s ranking. Delete X from those candidates’ preferences. Do “top-top” again. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 11/31
The French job market If “top-top” + “clean-the-lists” do not produce a complete matching: Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 12/31
The French job market If “top-top” + “clean-the-lists” do not produce a complete matching: 3. Roth and Sotomayor (1990), page 27 (with candidates-proposing). Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 12/31
The French job market If “top-top” + “clean-the-lists” do not produce a complete matching: 3. Roth and Sotomayor (1990), page 27 (with candidates-proposing). In the last 25 years, across all 74 sections, step 3 was activated about 10 times! Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 12/31
The French job market Quick summary: The job market is very “costly” Not fi lling a position is the worst outcome for the departments; Participants have little understanding of the assignment procedure (i.e., the algorithm). Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 13/31
The French mathematicians In 1998 a group of mathematicians set up a web-side: “opération postes”. It records in real-time: Job openings (with fi eld label) + list of candidates with the right to enter the market; interview schedules; candidates’ rankings by the departments. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 14/31
Our data Our data covers the period 1999-2007. It also includes: Year and origin of PhD of all ranked candidates (some candidates are interviewed but not ranked). All publications (as from 1990) by: all ranked candidates; all hiring institutions; → About 50’000 publications. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 15/31
Our data Publication records include: year; journal (matched to its impact factor); AMS (ordered) sub fi elds (in order of importance). (“JEL codes” for mathematicians) Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 16/31
Our data We also have: The fi nal match; But we do not have: Candidates’ preferences. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 17/31
Overview The French job market for (junior) mathematicians is about: Tenure positions. Low salary, high teaching load, little room for negociation. 100 positions per year (min = 73, max = 123). Candidates (per year): ≥ 500 on the market 330–450 candidates interviewed (average: 4 times the number of positions) 200–250 candiates ranked (about twice the number of positions). Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 18/31
Issue & Observation Departments are quite good at “targeting” their candidates (coordination?) Distribution of ranks of matched candidates 1 2 3 4 5 70 % 18 % 7.3% 2.9 % 1.8% Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 19/31
Issue & Observation In general, we need both sides’ preferences to compute the match. But in some cases we don’t: Candidate A ranked fi rst by some department X. Only department X ranked A. A’s only possible choice is whether A acceptable or not. If acceptable (always is), A takes the job at X. Candidates ranked by X below A are thus ranked one time less. Repeat. . . Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 20/31
Issue & Observation Positions fi lled without needing candidates’ preferences: year pos. fi lled % of all positions 1999 39 32.5 2000 45 50.5 2001 33 38.4 2002 40 57.1 2003 49 50.5 2004 36 49.3 2005 37 41.5 2006 41 33.6 2007 52 45.2 Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 21/31
Who’s the culprit? Candidates ranked fi rst at some date not ranked at a later date: Their obtained their most preferred department, no need to attend more interviews (and thus are not ranked). Departments consider them as matched, so do not rank them (or rank them in ineligible position). Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 22/31
Who’s the culprit? For each matched candidate, consider � interviews and � succesful interviews that happen after the interview with the hiring dept. All ranked 1st only Comp. sub-market .481 .458 Non-comp. sub-market .456 .457 In the non-comp. sub-market, departments are more likely to fi ll their ranking with “rubish.” Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 23/31
Sorting Ranking departments and candidates presents some hurdles: Impact factors can be “polluted” by non-pure math journals (e.g., Nature ≈ 10 × Annals of Mathematics ) ⇒ Consider sub fi elds. Most candidates do not have (or have too few) publications to obtain meaningful rankings. Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 24/31
Sorting Departements and candidates classi fi ed into quartiles. Ranking of departments is quite “stable” over the years, not too “sub fi elf sensitive.” For candidates, considered the publications at years y-2, y-1, y, y+1. Used the squares of impact factors. Outliers (i.e., publishing in Nature/Science/PNAS/. . . ) deleted (about 30 observations). Knocking out the Supplyand Sorting in Decentralized Job Markets – p. 25/31
Recommend
More recommend
