Using machine learning for communication classification

Penczynski, Stefan P.

doi:10.1007/s10683-018-09600-z

Cited by 20 publications

(17 citation statements)

References 22 publications

(31 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The use of coding software and machine learning techniques is less widespread for the analysis of natural language data from economic experiments. Penczynski (2016) is to our knowledge the first to investigate the suitability of machine learning for the classification of free-form communication in laboratory experiments and shows that the classification results are promising compared to human classifications. The messages are first broken down into tokens.…”

Section: Coding Software Machine Learningmentioning

confidence: 94%

“…Argumentation (or argument) mining is an interdisciplinary field of research that tries to extract arguments from unstructured textual documents using computational linguistic and machine learning advances (Lippi and Torroni 2016). In experimental economics, researchers have proposed coding methods using both humans (for instance, Cooper and Kagel 2005;Houser and Xiao 2011;Eich and Penczynski 2016) and computerized techniques like machine learning (for instance, Penczynski, 2016).…”

Section: Sciencesmentioning

confidence: 99%

“…Tokens are strings -usually words or numbers -separated by spaces. Tokens that are not relevant for the text content, such as stopwords, are removed and inflected relevant words are reduced to their stem (for instance, "team," "teams," and "teamed" are reduced to "team," Penczynski, 2016). The remaining tokens are counted and the rarely used tokens are removed.…”

Section: Coding Software Machine Learningmentioning

confidence: 99%

“…A substantial share of the dataset is used for training the model used by the learning method (in-sample), and the remaining minor share for testing the model (out-of-sample). As a result, certain words (for instance, the word "think") can be identified to be correlated with a particular behavior (for instance, the level of reasoning in the beauty-contest game, Penczynski, 2016). Coding practices in Economics for quasi-experimental (for instance, Freddi, 2017) and non-experimental (for instance, Gentzkow and Shapiro 2010;Hansen et al 2017;Gentzkow, Kelly, and Taddy, 2017;Gentzkow, Shapiro, and Taddy, 2017;Ash et al 2017) natural language data can provide guidance as to the use of software programs and machine learning techniques.…”

Section: Coding Software Machine Learningmentioning

confidence: 99%

“…The coding categories may be chosen such that the link to choices and outcomes in the experiment is meaningful. Meaning can be derived 13 For instance, Penczynski (2016) takes out tokens that appear less than five times in all messages together. from standard economic theory (e.g., reasoning, strategic thinking, equilibrium play, efficiency and payoff considerations), behavioral theories (e.g., other-regarding preferences, fairness, social norms, promises, social distance and identity), or in an exploratory way (e.g., observations, emotions, and humor).…”

Section: Causal Links Between Coding Categories and Outcomesmentioning

confidence: 99%

See 4 more Smart Citations

Communication in laboratory experiments

Brandts¹,

Cooper

Rott³

2019

Handbook of Research Methods and Applications in Experimental Economics

View full text Add to dashboard Cite

In this chapter we present a first survey of laboratory studies using communication. We first discuss a number of issues about the implementation of communication in the laboratory. In particular we discuss variations in the channels, the structure and the content analysis of communication. Second, we survey a number of studies in which the effects of free-form communication are compared to those with more restricted communication. We finish with a brief reference to the question why communication matters, to what the mechanisms through which communication affects behavior are. Table 1: Dimensions of communication 4 We do not cover communication in field experiments. CHANNELS OF COMMUNICATIONPossible channels of communication in laboratory experiments are paper-and-pencil messages, chat, audio, video, or face-to-face interactions. The following is an incomplete reference list. Paper-and-pencil (Charness and Dufwenberg 2006) and chat messages (Brandts, Charness, and Ellman, 2016;Brandts, Rott, and Sola, 2016) are sent in written form, whereas audio (Greiner et al. 2012), video (Brosig et al. 2003Greiner et al. 2012), and face-to-face messages (Isaac and Walker 1988;Bochet et al. 2006;Schram et al. forthcoming) are usually exchanged orally. In faceto-face communication, participants meet physically and verbally communicate with each other (for instance, Isaac and Walker, 1988).The main drawback of audio, video and face-to-face communication is that typically they do not allow for (complete) anonymity, which entails that the interaction between participants does not necessarily stop at the end of the experiment. 5 An additional limitation of face-to-face communication is the difficulty of creating a record of what has been said. By contrast, chat, audio, and video messages can be recorded, transmitted and analyzed at a later stage (for instance, to a subsequent generation as in Chaudhuri et al., 2009). a computerized experiment, a chat window opens (or participants have the option to open it) and participants can type in text messages. 6 STRUCTURE OF COMMUNICATIONWith the term 'structure' we refer here to the arrangements of who can communicate to whom, when, in what order, and how often. All these elements can be tightly controlled in the laboratory and this control allows for many potentially interesting variations. This makes it possible to study the general phenomenon of 'communication' in a very systematic way. We organize our discussion of structure into three subsections: network and direction, frequency/repetition, and order and timing. COMMUNICATION NETWORK AND DIRECTIONIn experiments involving only two participants the only issue is whether they both can communicate with the other participant or whether only one of the two can (and who it is). In experiments with more than two participants, there are even more possibilities, since now there are several options both with respect to who can talk to whom and who can listen to whom. The structure of these connections is what we refer to as communication netw...

show abstract

Section: Coding Software Machine Learningmentioning

confidence: 94%

Section: Sciencesmentioning

confidence: 99%

Section: Coding Software Machine Learningmentioning

confidence: 99%

Section: Coding Software Machine Learningmentioning

confidence: 99%

Section: Causal Links Between Coding Categories and Outcomesmentioning

confidence: 99%

See 3 more Smart Citations

Communication in laboratory experiments

Brandts¹,

Cooper

Rott³

2019

Handbook of Research Methods and Applications in Experimental Economics

View full text Add to dashboard Cite

show abstract

Testing for the emergence of spontaneous order

Georgalos

Hey

2019

Exp Econ

View full text Add to dashboard Cite

We report on an experimental investigation of the emergence of Spontaneous Order, the idea that societies can coordinate , without government intervention, on a form of society that is good for its citizens, as described by Adam Smith. Our experimental design is based on a production game with a convex input provision possibility frontier, where subjects have to choose a point on this frontier. We start with a simple society consisting of just two people, two inputs, one final good and in which the production process exhibits returns to specialisation. We then study more complex societies by increasing the size of the society (groups of 6 and 9 subjects) and the number of inputs (6 and 9 inputs respectively), as well as the combinations of inputs that each subject can provide. This form of production can be characterised as a cooperative game, where the Nash equilibrium predicts that the optimal outcome is achieved when each member of this society specialises in the provision of a single input. Based on this framework, we investigate whether Spontaneous Order can emerge, without it being imposed by the government. We find strong evidence in favour of the emergence of Spontaneous Order, with communication being an important factor. Using text classification algorithms (Multinomial Naive Bayes) we quantitatively analyse the available chat data and we provide insight into the kind of communication that fosters specialisation in the absence of external involvement. We note that, while communication has been shown to foster coordination in other contexts (for example, in public goods games, market entry games and competitive coordination games) this contribution is in the context of a production game where specialisation is crucial.

show abstract

Strategic environment effect and communication

Hanaki

Ozkes

2022

Exp Econ

View full text Add to dashboard Cite

We study the interaction of the effects of the strategic environment and communication on the observed levels of cooperation in two-person finitely repeated games with a Pareto-inefficient Nash equilibrium and replicate previous findings that point to higher levels of tacit cooperation under strategic complementarity than under strategic substitutability. We find that this is not because of differences in the levels of reciprocity as previously suggested. Instead, we demonstrate that slow learning coupled with noisy choices may drive this effect. When subjects are allowed to communicate in free-form online chats before making choices, cooperation levels increase significantly to the extent that the difference between strategic complements and substitutes disappears. A machine-assisted natural language processing approach then shows how the content of communication is dependent on the strategic environment and cooperative behavior, and indicates that subjects in complementarity games reach full cooperation by agreeing on gradual moves toward it.

show abstract

Using machine learning for communication classification

Cited by 20 publications

References 22 publications

Communication in laboratory experiments

Communication in laboratory experiments

Testing for the emergence of spontaneous order

Strategic environment effect and communication

Contact Info

Product

Resources

About