• No results found

The algorithm for generating networks

7.5 Empirical case

7.6.1 The algorithm for generating networks

The network is represented in the form of adjacency matrix 𝑋 with 𝑛 rows and 𝑛 columns, both corresponding to the number of nodes in the network. The links have values which are only considered to control the duration of the links. The local network mechanisms and the global network structures are analysed by considering binarized networks. A link in a binarized network exists if the value in a corresponding valued network is higher than 0.

175

The algorithm allows different initial networks to be to specified, namely either an empty network, random network or a network with a specific global network structure (e.g. a blockmodel). Besides initial network 𝑋, parameters πœ† and πœ… have to be set. Parameter πœ† expresses the maximum expected out-degree, while parameter πœ… relates to the number of iterations. The local network mechanisms must also be provided with the corresponding vector πœƒ which operationalizes the importance (strength) of the selected local network mechanisms.

The algorithm is iterative. At each iteration, one node (ego 𝑖) is selected among all the nodes in the network (each node is selected with equal probability)31. Considering 𝑖 and the selected local

network mechanisms, the network statistics are calculated and weighted by πœƒ. The weighted network statistics are normalized on the interval between 0 and 1. From among 25% of the nodes with the highest value of the weighted network statistics, one node is randomly selected.

In addition, the tenure is calculated at each iteration and new nodes (newcomers) are added to the network and some existing nodes are removed (outgoers) at selected iterations.

Weighted network statistics

The weighted network statistics are calculated by the function π‘π‘œπ‘šπ‘π‘’π‘‘π‘’. 𝑆(𝑋, 𝑖, 𝑀, πœƒ) that considers the set of mechanisms 𝑀 and the weights of the corresponding mechanisms πœƒ. 𝑀 is the set of operationalized mechanisms defined on the binarized network 𝑋 and node 𝑖.

The computed value of a given mechanism (from the set of mechanisms 𝑀) is a vector of length 𝑛. Each element of this vector corresponds to one node in the network. When several mechanisms are considered, the vectors can be organized into matrix 𝐻 with 𝑛 rows and π‘š columns representing the mechanisms. The matrix that is obtained is weighted as 𝑆 = π»πœƒπ‘‡, resulting in a

vector of length 𝑛 which is returned by the function π‘π‘œπ‘šπ‘π‘’π‘‘π‘’. 𝑆(𝑋, 𝑖, 𝑀, πœƒ).

31 The probabilities could vary among the nodes. For example, it could be assumed that those nodes with a lower tenure will have more opportunities to ask for advice. However, whether this is a reasonable assumption depends on the company’s policies and organizational culture. To consider the most parsimonious case, it is assumed in this study that all nodes have equal probabilities of asking for advice at any time.

176

Duration of the links

No specific mechanism that is considered in this study is able to control the duration of a link (i.e. duration of the interaction between the advice-seeker and advice-giver). Instead, it is assumed that all interactions last the same amount of time. One unit of time is defined through the number of iterations at which each node will get (in average) one opportunity to establish a link. The number of iterations depends on the size of the network and the desired maximum expected out-degree (parameter πœ†).

Let us consider the case without newcomers and outgoers Let us also assume there are 𝑛 nodes in the network and each node has up to πœ† opportunities to establish a link. On the assumption the nodes are chosen randomly with equal probability, the number of iterations needed to reach the expected number of opportunities to establish a link is 𝛼 = πœ†π‘› (this is the length of one unit of time). Over 𝛼 iterations, each node receives (on average) πœ† opportunities to establish a link. This implies that some individual nodes can have a higher out-degree, which happens because some receive more opportunities to establish a link than others. In a very unlikely case, each node could receive exactly πœ† opportunities to create a link. In that case, the maximum out-degree of each node is exactly π‘šπ‘–π‘›(𝑛, πœ†) if loops are allowed and if no tie is chosen twice (in a sense, confirmed or reset).

Links last a limited amount of time. Specifically, the duration of links is set to 𝛼 + 1 iterations. When new nodes are added to the network, parameter 𝛼 must be updated by considering the new number of nodes. This implies the number of iterations between different waves can vary. The algorithm is implemented in such a way that the number of outgoers does not affect the number of iterations.

In order to ensure sufficient iterations so that the mechanisms being considered can affect the global network structure considerably, the number of iterations is multiplied by the constant πœ…. The value of πœ… > 1 increases the expected number of opportunities for each node to establish a link while it affects neither a link’s duration nor the (maximal) expected number of links. A higher expected number of opportunities for each node to establish a tie can also makes the structure more stable before the new nodes are added. In other words, a higher number of iterations gives β€œmore time” to the mechanisms to affect the global network structure.

177

Newcomers

New nodes can be added over one or several waves. The iterations in which new nodes are added to the network can be selected in different ways: (i) a single node can be added at a time; or (ii) several nodes can be added at once. Further, the node (or several nodes) can be added at randomly selected iterations or at predefined iterations, e.g. equally distributed across iterations. In this study, newcomers are added in three waves. The number of newcomers for each wave is represented by vector β„΅. The number of iterations between each wave is determined according to the total number of nodes in the network, based on parameter πœ† and parameter πœ….

Algorithm 7.1: The algorithm for generating networks

import initial network 𝑋 (a matrix with 𝑛 rows and 𝑛 columns, where 𝑛 is the number of units)

import πœƒ (a vector with the mechanisms’ weights)

import 𝑀 (a set of functions which defines the mechanisms) set πœ† (the expected maximum out-degree)

set πœ… (the factor by which the number of iterations must be increased between the waves)

set β„΅ (a vector with the number of newcomers per waves)

set 𝑂 (a vector with iterations at which the outgoings are to be removed) set 𝑇 (tenure, a vector of length 𝑛)

compute π‘“π‘œπ‘Ÿπ‘‡π‘’π‘›π‘’π‘Ÿπ‘’πΆπ‘œπ‘Ÿπ‘‰π‘’π‘ = cumsum(℡𝐴) βˆ— πœ† βˆ— πœ… (the number of iterations between consecutive

waves of newcomers)

compute 𝐸 = cumsum(π‘“π‘œπ‘Ÿπ‘‡π‘’π‘›π‘’π‘Ÿπ‘’πΆπ‘œπ‘Ÿπ‘‰π‘’π‘) (iterations at which the newcomers are added to the network)

compute π‘˜ = max(𝐸) (the total number of iterations) set π‘“π‘œπ‘Ÿπ‘‡π‘’π‘›π‘’π‘Ÿπ‘’πΆπ‘œπ‘Ÿ to first element of π‘“π‘œπ‘Ÿπ‘‡π‘’π‘›π‘’π‘Ÿπ‘’πΆπ‘œπ‘Ÿπ‘‰π‘’π‘ for 𝑙 in 1:π‘˜

|__| set 𝑇 = 𝑇 + 1/π‘“π‘œπ‘Ÿπ‘‡π‘’π‘›π‘’π‘Ÿπ‘’πΆπ‘œπ‘Ÿ

|__| randomly select a unit 𝑖 (actor/ego)

|__| calculate 𝑆 = π‘π‘œπ‘šπ‘π‘’π‘‘π‘’. 𝑆(𝑋𝑙, 𝑖, 𝑀, πœƒ) (a vector of the weighted network statistics with

the length 𝑛)

|__| calculate πœ‘ = π‘†βˆ’min (𝑆)

max(𝑆)βˆ’min (𝑆) (normalize S, so the min(𝑆) = 0 and max(𝑆) = 1)

|__| if πœ‘ β‰₯ 𝑄3(πœ‘) then classify a corresponding unit into set 𝐢 (where 𝑄3 is the third quartile)

|__| randomly select unit 𝑗 among the units from set 𝐢 |__| set a link 𝑖 β†’ 𝑗

|__| calculate 𝑋 = 𝑋 βˆ’ 1 (πœ†π‘› + 1)⁄ |__| calculate 𝑋 = {0, π‘₯ ≀ 0

π‘₯, π‘₯ > 0 | if 𝑙 ∈ 𝑂

|________|randomly select a unit or a group of units to be removed |________|remove the selected unit(s) and update X and T accordingly |if 𝑙 ∈ 𝐸 and 𝑙 β‰  π‘˜:

|________|add a unit or a group of units and update 𝑋 and 𝑇 accordingly |________|set π‘“π‘œπ‘Ÿπ‘‡π‘’π‘›π‘’π‘Ÿπ‘’πΆπ‘œπ‘Ÿ to next element of π‘“π‘œπ‘Ÿπ‘‡π‘’π‘›π‘’π‘Ÿπ‘’πΆπ‘œπ‘Ÿπ‘‰π‘’π‘

178

Outgoers

The number of outgoers can be selected arbitrarily. They can leave the network in waves just before or after newcomers are added or can leave the network one by one. In this implementation of the algorithm, outgoers leave the network at the selected iterations which are in vector 𝑂. The nodes to be removed from the network can be selected based on their personal characteristics (e.g. tenure), network characteristics (e.g. popularity or hierarchical level) or randomly. Here, the nodes to be removed are selected randomly, which is in line with observations from the empirical data. The number of nodes to be removed from the network is 25% of all the nodes in the network, calculated immediately after a wave of newcomers has been added to the network.