## Abstract

This paper reports results of a simulation designed to evaluate the precision of Neumann approximations of outcomes of networked economic systems. We simulate systems with Erdös-Renyi, Watts-Strogatz, and Barabási-Albert networks. We discuss the conditions under which second-order approximations of economic outcomes generate small errors. We find that in certain economic systems data requirement is significantly reduced if the research goal is to predict economic outcomes of targeted agents (as opposed to outcomes of the entire system). Despite the systems’ complex network structure, our simulations indicate that sampling the targeted group, its connections, and the connections of its connections is sufficient to predict outcomes. As a result, economic policy that targets this specific group should focus on agents in the two-order of distance sample as opposed to the entire network. This sample of agents can be thought of as the group’s agents of change or opinion leaders.

We’re sorry, something doesn't seem to be working properly.

Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.

## Notes

Neumann approximations are formally defined in Section 3.

The peer-elasticity

*b*indicates the weight agents place on their social networks, hence, it can be thought of as the strength of the system’s network effect. The social multiplier is computed as follows:*η*= 1/(1-*b*). Refer to Section 2 for a discussion.We impose

*w*_{ i i }=0 indicating that an agent is not connected to herself. Notice that**W**is not assumed to be symmetric.Notice that the row sums of

*b***A**are equal to 0 <*b*< 1, and the average column sum is also*b*.The effect of the connections of all

*i*’s connections is therefore captured by the*i*-th row of (*b***A**)^{2}.Network density is defined as the number of existing connections divided by the number of total possible connections.

Intransitivity is the the probability that adjacent agents of an agent are not connected. Thus, high intransitivity implies low clustering.

Note that

*k*_{ i }correspond to the number of adjacent links of agent*i*which were not initiated by*i*itself.The power law generates degree distributions with positive skew.

225 models correspond to: 9 (number of

*b*s) × 9 (number of Erdös-Renyi networks) + 9 (number of*b*s) × 8 (number of Watts-Strogatz networks) + 9 (number of*b*s) × 8 (number of Barabási-Albert networks).Notice that the only random component of each model is the matrix

**A**. For each model, the simulation has the potential to generate 500 different outcome profiles based on 500 different matrices**A**. The data generated are 112,500 vectors**Y**(225 models × 500 replications) and 3,375,000 (112,500 × 30) approximations \(\hat {\mathbf {Y}}^{(\mathbf {k})}\).Since approximation errors are always non-negative, there is no reason to compute squared errors.

See rows of Table 1.

See rows of Table 2.

Approximated using the middle order of 1.5, i.e. −2.36=−2.45+0.06*(1.5).

Approximated using the middle order of 29.5, i.e. -0.68 = -2.45 + 0.06*(29.5).

**W**^{(S)}has (*n*−*n*^{S}) rows of zeros. The data of the non-zero rows of**W**^{(S)}correspond to the second-order snowball sample of the targeted group. Refer to Wasserman and Faust (1994, Chapter 2) for a detailed discussion of network data collection.The outcome variance among

*n*^{G}agents may be very different from the variance among*n*agents, especially when elements of**X**are not uniformly distributed. Distributions with a bell shape (mass point in the middle range of**X**) and a U-shape (mass point on low and high ranges of**X**) are studied in Section 6.All realizations

*x*_{ i }were positive.Data was obtained from the UCINET project’s website. Available at: https://sites.google.com/site/ucinetsoftware/datasets. Four inmates reported no connections. To comply with the assumptions of the model we randomly assigned one connection to each one of these inmates.

Data was obtained from Mathew Jackson’s webpage. Available at: http://www.stanford.edu/jacksonm/Data.html. An adjacent matrix was constructed from the matrix with number of citations assuming that a connection exists if there are positive citations. The diagonal, i.e. self-citation, was converted to zero.

For example, (Duflo and Saez 2002) investigate the participation decisions in a retirement plan and estimate a peer effect parameter of approximately 0.2, i.e. social multiplier of 1.25. Lin (2010) estimates peer effects in student academic achievement and finds a social multiplier of approximately 1.38. Anselin et al. (2010) study spatial effects with a spatial hedonic model to value the access to potable water and estimate a spatial multiplier of 1.32. Baller et al. (2001) estimate spatial effects for the homicide rates of southern US counties and find spatial multipliers of 1.22 and 1.30 for 1980 and 1990, respectively.

See Neilson and Wichmann (2014) for a discussion about network centrality and social welfare.

## References

Anselin L (2001) Spatial econometrics. A companion to theoretical econometrics

Anselin L (2003) Spatial externalities, spatial multipliers, and spatial econometrics. Int Reg Sci Rev 26(2):153–166

Anselin L, Lozano-Gracia N, Deichmann U, Lall S (2010) Valuing access to water: a spatial hedonic approach, with an application to bangalore, india. Spat Econ Anal 5(2):161–179

Baller RD, Anselin L, Messner SF, Deane G, Hawkins DF (2001) Structural covariates of us county homicide rates: incorporating spatial effects. Criminology 39(3):561–588

Barabási AL, Albert R (1999) Emergence of scaling in random networks. Science 286(5439):509–512

Bramoullé Y (2001) Interdependent utilities, preference indeterminacy, and social networks.

*Université de Toulouse,[mimeo]*Bramoullé Y, Djebbari H, Fortin B (2009) Identification of peer effects through social networks. J Econ 150(1):41–55

Choromański K, Matuszak M, Mikisz J (2013) Scale-free graph with preferential attachment and evolving internal vertex structure. J Stat Phys 151(6):1175–1183

Conley T, Udry C (2001) Social learning through networks: the adoption of new agricultural technologies in ghana. Am J Agric Econ 83(3):668–673

Copic J, Jackson MO, Kirman A (2009) Identifying community structures from network data via maximum likelihood methods. BE J Theor Econ 9(1)

Costenbader E, Valente TW (2003) The stability of centrality measures when networks are sampled. Soc Netw 25(4):283–307

Duflo E, Saez E (2002) Participation and investment decisions in a retirement plan: the influence of colleagues choices. J Publ Econ 85(1):121–148

Erdös P, Renyi A (1959) On random graphs. Publ Math 6:290–297

Erickson BH, Nosanchuk TA (1983) Applied network sampling. Soc Netw 5(4):367–382

Frank O (1979) Sampling and estimation in large social networks. Soc Netw 1(1):91–101

Galaskiewicz J (1991) Estimating point centrality using different network sampling techniques. Soc Netw 13(4):347–386

Glaeser EL, Sacerdote BI, Scheinkman JA (2003) The social multiplier. J Eur Econ Assoc 1(2-3):345–353

Illenberger J, Flötteröd G (2012) Estimating network properties from snowball sampled data. Soc Netw 34(4):701–711

Katz L (1953) A new status index derived from sociometric analysis. Psychometrika 18(1):39–43

Kelejian H, Prucha I (1998) A generalized spatial two-stage least squares procedure for estimating a spatial autoregressive model with autoregressive disturbances. J Real Estate Financ Econ 17(1):99–121

Kelejian HH, Prucha IR, Yuzefovich Y (2004) Instrumental variable estimation of a spatial autoregressive model with autoregressive disturbances: large and small sample results. Adv Econ 18:163–198

Lam S, Schaubroeck J (2000) A field experiment testing frontline opinion leaders as change agents. J Appl Psychol 85 (6):987

Laschever R (2013) Keeping up with ceo jones: Benchmarking and executive compensation. J Econ Behav Organ 93:78-100

Lee L (2003) Best spatial two-stage least squares estimators for a spatial autoregressive model with autoregressive disturbances. Econ Rev 22(4):307–335

Lin X (2010) Identifying peer effects in student academic achievement by spatial autoregressive models with group unobservables. J Labor Econ 28(4):825–860

MacRae D (1960) Direct factor analysis of sociometric data. Sociometry 23(4):360–371

Manski CF (1993) Identification of endogenous social effects: the reflection problem. Rev Econ Stud 60(3):531–542

Meyer C (2000) Matrix analysis and applied linear algebra, Volume 2. Society for Industrial and Applied Mathematics

Monge M, Hartwick F, Halgin D (2008) How change agents and social capital influence the adoption of innovations among small farmers. Technical report, IFPRI Discussion Paper 00761. Washington, DC: IFPRI

Nair H, Manchanda P, Bhatia T (2010) Asymmetric social interactions in physician prescription behavior: The role of opinion leaders. J Mark Res 47(5):883–895

Neilson W, Wichmann B (2014) Social networks and non-market valuations. J Environ Econ Manag 67(2):155–170

Neilson W, Winter H (2008) Votes based on protracted deliberations. J Econ Behav Org 67(1):308–321

Onnela JP, Saramäki J, Hyvönen J, Szabó G, Lazer D, Kaski K, Kertész J, Barabási AL (2007) Structure and tie strengths in mobile communication networks. Proc Natl Acad Sci 104(18):7332–7336

Taussky O (1949) A recurring theorem on determinants. Am Math Mon 56(10):672–676

Wasserman S, Faust K (1994) Social Network Analysis: Methods and Applications. Cambridge University Press, Cambridge

Watts DJ, Strogatz SH (1998) Collective dynamics of small-world networks. Nature 393(6684):440–442

## Author information

### Authors and Affiliations

### Corresponding author

## Rights and permissions

## About this article

### Cite this article

Wichmann, B. Agents of Change and the Approximation of Network Outcomes: a Simulation Study.
*Netw Spat Econ* **15**, 17–41 (2015). https://doi.org/10.1007/s11067-014-9266-2

Published:

Issue Date:

DOI: https://doi.org/10.1007/s11067-014-9266-2