Research process

We constrained our research focus to the impact of individual overcondence in infor-mation markets and, as illustrated in the previous section, the experimental methods chosen appear to be most salient for studying the individual and group-based level of analysis. Figure 5.1 reects our research process for the empirical studies. We began at the most granular level by testing whether condence in innovation evaluation could be experimentally manipulated as an individual trait. After developing a feasible manipu-lation to create articial condence levels in the laboratory, we continued on to the rst experiment addressing the relationship between condence and participation behavior in the context of information markets. Here, the focus was set upon individual behavior in a controlled market environment to isolate the eects of condence. Finally, we studied the interaction between individual participation behavior at controlled condence levels in information markets and other human agents. In this experiment, the relationship between participants' condence levels and information market evaluation quality be-came the focus of attention.

In the following subsections, we will introduce the three steps of the research process.

Each subsection will briey address the theoretical and empirical background for our study and highlight our specic goals with regard to the gaps in the current literature and our central research questions.

Figure 5.1.: Research process and level of analysis of the three empirical studies (Source:

Own depiction)

5.4.1. Individual condence manipulation

Causal relationships are studied in laboratory experiments by examining the eects that result from manipulating a particular impact factor. In our case, we aimed to study the impact of dierent condence levels.

Drawing on research in social psychology, there are numerous examples where exper-imenters have successfully shaped participants' self-concepts via false-feedback ma-nipulation.

Frey (1978) provides an early example in which personal performance is experimentally manipulated to test subjects' reactions to success and failure in public or private con-ditions after receiving or evaluating performance feedback. In the experiment, student subjects took an intelligence test and received ctitious results that were either below or above the average. Here, the performance-feedback manipulation induced signicant dierences in subjects' subsequent behavior. Depending on their perceived performance and public or private feedback on their test scores, the subjects systematically chose dierent strategies to increase presented self-value.

More closely related to self-concept manipulation, Greenberg et al. (1992) articially

induced high and low self-esteem via false feedback. These experiments focused on the relationship between self esteem and anxiety levels in experiencing and expecting shock.

The authors found that student subjects manipulated to have high self-esteem physio-logically experienced and reported signicantly lower levels of anxiety when experiencing or expecting shock.

The specic manipulation of overcondence has only recently been applied in a psychological experiment. Anderson et al. (2012) studied the status-enhancing eect of overcondence. The researchers used overly positive feedback on a pre-experimental estimation task to induce overcondence. This treatment successfully induced overcon-dence in student subjects while preventing subjects from gaining increased self-esteem or suspecting that they had been given false feedback. After the manipulation, overcon-dent subjects were paired with subjects who did not receive the treatment to repeat a similar task as a team. The study showed that articially-induced overcondence positively inuences subjects' perceived competency and status, both by themselves and by their respective partners. However, the researchers stress that the condence manipulation carried out specically focused on the ability to predict other peoples' characteristics. They highlight that individuals are often largely unaware of their accu-racy in perceiving others, which makes it easier to exploit in experimental manipulation (Ames and Kammrath 2004). While their study gives indication that condence can be successfully manipulated in a laboratory, it lacks relation to overcondence in the context of innovation evaluation.

In summary, a large body of experimental research in psychology has demonstrated the feasibility and success of manipulating self-concepts in general and overcondence in par-ticular. Based on the research, it is possible to develop and test a suitable treatment that allows us to induce overcondence by providing manipulated feedback. The treatment check addresses the domain of innovation evaluation because we later focus on innovation evaluation tasks. Also, the above-mentioned study by Anderson et al. (2012) highlighted that the specic manipulation they used should not be indiscriminately transferred to dierent task domains.

5.4.2. Overcondence and individual behavior in information markets

On the individual level, we aimed to gain an understanding of the impact of overcon-dence on individual behavior in information markets. However, in real-life information

markets with multiple human participants, observing individual traits and overall mar-ket behavior to help understand relationships between individual traits and individual behavior may be impossible. All market actions (except the rst market action) are potentially subject to interactions between the initial subjective condition and the mar-ket actions of other participants. This would likely prevent a researcher from validly explain variance in a subject's behavior based on his condition alone, which, in our case, relates to condence. Yet, researchers have also stressed that too little is known about the direct relationship between condence and market behavior (Oberlechner and Osler 2012). Our rst experiment therefore set out to create an experimental environment that would allow us to learn about the direct eects of dierent condence levels on individual trading behavior in information markets.

In an experimental setting, a researcher should rigorously control the environment across dierent treatment conditions. In a market experiment, such control requires stability of the market environment across treatments. Researchers have previously engaged actors to control and stabilize live human interactions in studies that focused on economic be-havior (e.g. Kopelman et al. (2006)). In an information market, such actors could aim to keep their trading stable based on pre-dened rules that are remain constant over ex-perimental conditions. Yet, information market experiments do not necessarily require the visual presence of other human participants. The human actor may be exchangeable with an articial agent if it acts suciently human. Earlier research has pointed to the importance of human subjects at least perceiving the agents to be human. Otherwise, they could exhibit behavior that might be more geared toward interaction with articial machines, e.g. appearing to be less disciplined in their actions (Brown-Kruse 1991).

Yet, more recent experiments in market economics have provided evidence that articial agents can be incorporated in human-subject experiments without provoking unnatural human-subject behavior (Duy 2006). We drew from these ndings to create an arti-cial market environment for the rst experiment that would best control the impact of overcondence on individual behavior. In the experiment, human subjects would per-ceive that they were trading with other human subjects, while really interacting with an articial market maker.

We integrated the ndings from Section 4.3 and Section 4.4 to develop our hypotheses about the relationship between overcondence and individual behavior in information markets for innovation evaluation. The hypotheses will be presented at the beginning of Chapter 7.

5.4.3. Overcondence and the prediction quality of information markets

The experimental design of the rst experiment prevented us from drawing inferences about the impact of overcondence on the prediction quality of information markets.

However, prediction error is arguably the most important success variable in the context of information markets for innovation evaluation. Initiators of information markets will be more interested to learn how to deal with excessive condence if it signicantly alters the predictive quality of the markets. Hence, we developed a second experiment, pre-sented in Chapter 8, that allowed us to study the impact of overcondence on prediction quality as a group-based outcome of information markets. We integrated the theoretical ndings from Section 4.4 about the potential impact of overcondence on the predic-tion quality of informapredic-tion markets for innovapredic-tion evaluapredic-tion and also drew from the empirical ndings that will be reported in Chapter 7. Here, particular attention will be given to the dimensions of increased aggressiveness in trading by overcondent subjects and the contingency eects of this aggressiveness on non-treated subjects' trading and evaluation behavior.

Furthermore, Section 2.2 documented that innovation projects likely induce dierent be-haviors for searching out and acquiring diagnostic information, depending on how much information is initially available to the evaluators and the price to search for additional information.

We took these earlier ndings into account by studying the impact of overcondence on the prediction quality of information markets in two scenarios in which information market participants have distinct access to diagnostic information.

First, we created a basic experimental setting in which we controlled how well subjects were informed regarding the prediction task. All subjects received free pieces of similar diagnostic information, which was aimed at creating similar knowledge about the predic-tion task among the subjects. The subjects were expected to use the given informapredic-tion to form expectations regarding the prediction task and to trade within the information market. In this basic setting, we began by treating a set of subjects either with over-condence or low over-condence. Next, we awarded them with diagnostic information, and brought them together with a matching number of uninformed traders, who received nei-ther manipulated feedback nor diagnostic information. Lastly, we compared the impact of overcondent traders compared to low-condence traders on the prediction quality of the information markets in the case where treated subjects were given free access to

diagnostic information.

In addition, we investigated how overcondence impacts the willingness of subjects to engage in costly information searching. For example, entrepreneurship scholars have shown that overcondent individuals are less likely to engage in information acquisition or search to make more reasoned decisions (Trevelyan 2008). As a consequence, higher condence may negatively impact information seeking in innovation endeavors where uncertainty is particularly high. Again, a group of subjects were either treated with overcondence or low condence. Then, instead of receiving information for free, sub-jects were given the opportunity to acquire information for a cost after learning about the information markets' underlying prediction task. As in the basic experimental set-ting, treated subjects then entered the information market together with a matching number of uninformed traders, who received neither manipulated feedback nor diagnos-tic information. Lastly, we assessed the impact of overcondent traders compared to low-condence traders on the prediction quality of the information market in the case where the treated subjects were given the opportunity to acquire diagnostic information at a cost.