Presentation is loading. Please wait.

Presentation is loading. Please wait.

CREST-ENSAE Mini-course Microeconometrics of Modeling Labor Markets Using Linked Employer-Employee Data John M. Abowd portions of today’s lecture are the.

Similar presentations


Presentation on theme: "CREST-ENSAE Mini-course Microeconometrics of Modeling Labor Markets Using Linked Employer-Employee Data John M. Abowd portions of today’s lecture are the."— Presentation transcript:

1 CREST-ENSAE Mini-course Microeconometrics of Modeling Labor Markets Using Linked Employer-Employee Data John M. Abowd portions of today’s lecture are the work of Kevin McKinney (U.S. Census Bureau) and Ian Schmutte (University of Georgia) June 6, 2013

2 Topics May 30: Basics of analyzing complex linked data June 3: Basics of graph theory with applications to labor markets June 6: Matching and sorting models June 10: Endogenous mobility models Online course materials 6 June 2013© John M. Abowd and others, 20132

3 Lecture 3 More results on graph sampling More results on using modularity to find communities Matching and sorting models 6 June 2013© John M. Abowd and others, 20133

4 MORE APPLICATIONS OF SAMPLING GRAPHS (WORK OF MCKINNEY) 6 June 2013© John M. Abowd and others, 20134

5 Generating Connected Samples Let B be the upper right-hand corner of the adjacency matrix of the realized mobility network, as defined in lecture 2 B has I rows (one for each worker) B has J columns (one for each employer) Elements of B are the counts of the number of periods that worker i was employed by employer j 6 June 2013© John M. Abowd and others, 20135

6 Projection onto the Employer Nodes The projection of the bipartite graph of the realized mobility network onto employer nodes is accomplished by computing a new adjacency matrix P F = B T B Let S F be the column sums of B (employer degree distribution Then the adjacency matrix of the employer projection is AF = 1(P F - diag(S F )>0), where 1() is the element-wise indicator function 6 June 2013© John M. Abowd and others, 20136

7 Generating Connected Samples Accomplished by taking a sample of nodes from the employer projection where each node has a constant steady state selection probability of 1/(I – J) rather than the usual random walk procedure where each node would have selection probability of degree(node j )/(2|P F |) where |P F | is the number of edges in the employer projection 6 June 2013© John M. Abowd and others, 20137

8 6 June 2013© John M. Abowd and others, 20138

9 6 June 2013© John M. Abowd and others, 20139

10 6 June 2013© John M. Abowd and others, 201310

11 6 June 2013© John M. Abowd and others, 201311

12 6 June 2013© John M. Abowd and others, 201312

13 6 June 2013© John M. Abowd and others, 201313

14 MORE APPLICATIONS OF MODULARITY MODELING (WORK OF SCHMUTTE) 6 June 2013© John M. Abowd and others, 201314

15 6 June 2013© John M. Abowd and others, 201315

16 6 June 2013© John M. Abowd and others, 201316

17 6 June 2013© John M. Abowd and others, 201317

18 Modularity Determined Communities in Brazilian Linked EE Data RAIS graphs display the projection of the RMN in RAIS data (2002-2010) onto plants. After the projection, take a 5 percent random sample of plants along with their associated edges. The graph shows the structure after eliminating small degree nodes (degree <= 10). This shows about 25 percent of edges. Colors in the graph correspond to the maximum modularity partition of nodes into 15 communities at modularity of 0.65, which is very strong 6 June 2013© John M. Abowd and others, 201318

19 PSID Communities (Industry x Occupation Pseudo-employers) 6 June 2013© John M. Abowd and others, 201319

20 Raw Graph of Brazilian Data 6 June 2013© John M. Abowd and others, 201320

21 Maximum Modularity Communities in Brazilian EE Data 6 June 2013© John M. Abowd and others, 201321

22 Other Applications of Coloring Different definitions of communities (minimum cuts, maximum likelihood clustering) Showing conditional independence – Conditional independence can be used to massively increase the parallelization of a particular computation – This permits much faster estimation 6 June 2013© John M. Abowd and others, 201322

23 6 June 2013© John M. Abowd and others, 201323

24 6 June 2013© John M. Abowd and others, 201324

25 6 June 2013© John M. Abowd and others, 201325

26 6 June 2013© John M. Abowd and others, 201326

27 SORTING AND MATCHING MODELS 6 June 2013© John M. Abowd and others, 201327

28 How to Fit Matching Models Setting up equilibrium matching models Example using the Shimer (2005) model – Lecture based on Abowd, Kramarz, Perez-Duarte, and Schmutte (2012)Abowd, Kramarz, Perez-Duarte, and Schmutte (2012) Some critiques of these approaches 6 June 2013© John M. Abowd and others, 201328

29 6 June 2013© John M. Abowd and others, 201329

30 6 June 2013© John M. Abowd and others, 201330

31 6 June 2013© John M. Abowd and others, 201331

32 6 June 2013© John M. Abowd and others, 201332

33 6 June 2013© John M. Abowd and others, 201333

34 6 June 2013© John M. Abowd and others, 201334

35 6 June 2013© John M. Abowd and others, 201335

36 6 June 2013© John M. Abowd and others, 201336

37 6 June 2013© John M. Abowd and others, 201337

38 6 June 2013© John M. Abowd and others, 201338

39 6 June 2013© John M. Abowd and others, 201339

40 6 June 2013© John M. Abowd and others, 201340

41 6 June 2013© John M. Abowd and others, 201341

42 6 June 2013© John M. Abowd and others, 201342

43 6 June 2013© John M. Abowd and others, 201343

44 6 June 2013© John M. Abowd and others, 201344

45 6 June 2013© John M. Abowd and others, 201345

46 6 June 2013© John M. Abowd and others, 201346

47 6 June 2013© John M. Abowd and others, 201347

48 6 June 2013© John M. Abowd and others, 201348

49 6 June 2013© John M. Abowd and others, 201349

50 6 June 2013© John M. Abowd and others, 201350

51 6 June 2013© John M. Abowd and others, 201351

52 6 June 2013© John M. Abowd and others, 201352

53 6 June 2013© John M. Abowd and others, 201353

54 6 June 2013© John M. Abowd and others, 201354

55 6 June 2013© John M. Abowd and others, 201355

56 6 June 2013© John M. Abowd and others, 201356

57 6 June 2013© John M. Abowd and others, 201357

58 6 June 2013© John M. Abowd and others, 201358

59 6 June 2013© John M. Abowd and others, 201359

60 6 June 2013© John M. Abowd and others, 201360

61 6 June 2013© John M. Abowd and others, 201361

62 6 June 2013© John M. Abowd and others, 201362

63 6 June 2013© John M. Abowd and others, 201363

64 6 June 2013© John M. Abowd and others, 201364

65 6 June 2013© John M. Abowd and others, 201365

66 6 June 2013© John M. Abowd and others, 201366

67 6 June 2013© John M. Abowd and others, 201367

68 6 June 2013© John M. Abowd and others, 201368

69 6 June 2013© John M. Abowd and others, 201369

70 The Critique of Eeckhout and Kircher See Eeckhout and Kircher (2011)Eeckhout and Kircher (2011) Without frictions, their proposition 1 states that for any production function that can induce positive sorting there exists a production function that can induce negative sorting with the same equilibrium wage function 6 June 2013© John M. Abowd and others, 201370

71 The Critique of Eeckhout and Kircher See Eeckhout and Kircher (2011)Eeckhout and Kircher (2011) With search frictions, for every supermodular production function (production complements) that induces positive sorting, there exists a submodular production function (production substitutes) that induces the same wage distribution once the firm types are relabeled (reversing their order) 6 June 2013© John M. Abowd and others, 201371

72 What to Do? First, these points were made in the Abowd and Kramarz handbook paper (1999) regarding interpretations of the AKM decomposition as structural1999 Eeckhout and Kircher are much more general Must use additional outcomes – Actual vacancy data – Actual production data 6 June 2013© John M. Abowd and others, 201372

73 Summary Sampling matters for estimation of decomposition regardless of interpretation Maximum modularity methods produce interesting measures of the mobility communities in linked EE data Structural models of matching and sorting are feasible but the interpretation of the results may not be general 6 June 2013© John M. Abowd and others, 201373


Download ppt "CREST-ENSAE Mini-course Microeconometrics of Modeling Labor Markets Using Linked Employer-Employee Data John M. Abowd portions of today’s lecture are the."

Similar presentations


Ads by Google