Given The Provided Data, What Are Three Conclusions We Can Draw About Kickstarter Campaigns?

Abstract

The spread of information, opinions, preferences, and behavior across social media is a crucial feature of the current functioning of our economy, politics, and civilization. One of the emerging channels for spreading social collective action and funding of novelty in all these domains is Crowdfunding on various platforms such as Kickstarter, Indiegogo, Sellaband, and may others. The exact spreading mechanism of this commonage action is not well-understood. The general belief is that virality plays a crucial office. Namely, the mutual hypothesis is that the information or behavior propagates through individuals affecting 1 some other, presumably, through the links connecting them in social networks. The aim of our written report is to observe out the actual spreading mechanism in one particular case: spread of fiscal support for private Kickstarter campaigns. To our surprise, our studies show that "virality" plays here only a minor role. Nosotros used this outcome to construct a simple behavior-grounded stochastic predictor of the success of Kickstarter campaigns which is not based on the viral mechanism. The crucial feature of the model underlying the prediction algorithm is that the success of a campaign depends less on the backers influencing one another ("virality") but rather on the campaign appealing to a particular class of high-pledge backers. This entreatment is usually revealed at the very beginning of the campaign and it is an excellent success predictor. The example of Kickstarter is consistent with a recently proposed generic hypothesis that popularity in social media arises more from independent responses by individuals belonging to a big homophily class rather than from percolation, self-exciting processes, and other cooperative mechanisms resulting from common influence betwixt individuals. Thus, the very concept of "virality", which implies contagion betwixt participating individuals, plays just a minor function in the success machinery proposed hereby. A more appropriate term for the machinery underlying the social success in our model could exist "social appeal" or "social fettle".

Introduction and primal findings

Predicting collective human behavior is a very difficult task because the causes at the individual level (reciprocal influences, groups of individuals with similar behavior) are oft not directly recognizable from the systemic outcome.

Previous works uncovered a serial of feedback mechanisms amplifying "microscopic"/private inputs to the level of systemic transformations: multiplicative dynamics (Levy and Solomon, 1996), social percolation (Solomon et al., 2000), herding (Levy et al., 1994). Surprisingly, our findings for Kickstarter campaigns do not brandish such effects. We found that the success of a Kickstarter campaign depends on the arousal of a special type of backers' behavior which can be inferred from the analysis of the statistical distribution of pledges. In Fig. 1 we show representative pledge trajectories for failed and successful Kickstarter campaigns, as a hint to the data features which we apply for predicting the entrada outcome.

We measured the distribution of pledge sizes and compared the statistics of pledges for successful and failed campaigns. From the very first solar day, the successful campaigns display a "fat tail" distribution which follows a power-law dependence (direct line in the double-logarithmic plot of Fig. 2a). Conversely, the failed campaigns display much narrower exponential pledge distributions (directly lines in the semi-logarithmic plot of Fig. 2b).

One can attribute this huge difference in the shape of distributions to the backers' behavior, equally follows:

-In all campaigns there is a body of "rational" backers that pledge some "reasonable" fraction of the target sum leaving to the community of other backers to uphold the campaign. The latter have an exponential distribution of pledges characterized by a well-defined calibration (up to 2% of the target sum per day [come across Fig. 2]). Withal, the total pledge accumulated by this group of backers is mostly insufficient to achieve the target sum in a typical xxx-twenty-four hours campaign. Nosotros call this grouping "backers of blazon I".
-Successful campaigns, in addition to the "backers of type I", have a pocket-size group of backers who are especially fraternal to the campaign idea and react differently and independently of "reasonable proportion". We call these "backers of type II". Their "across objective/reasonable evaluation criteria" beliefs, originating in subjective "enthusiasm"/ "infatuation" with the project, leads to disproportionately loftier pledges.

Consequently, the distribution of pledges for successful campaigns spreads over a broad scale which is expressed in the power-law statistical distribution of pledges (Fig. 2). Notably, this power-law distribution is formed already in the first twenty-four hour period of entrada. The early appearance of the power-law pledge distribution rules out the mechanisms that involve the influence of the previous pledges on the electric current ones, every bit information technology was the case in (Levy and Solomon, 1996; Levy et al., 1994).

One may already extract more than general accept-home insights from these findings:

-For a entrada to succeed it is usually non enough to take a community of backers reasonably interested in information technology, it is much more of import to accept a core of enthusiastic backers committed to contribute more than than their "fair share" in the campaign goal. In futurity inquiry one may endeavor to find out which campaign characteristics evoke this kind of behavior.
-Our data suggest that unlike "viral propagation" whereby connected individuals transmit their enthusiasm ane to some other, the success of a Kickstarter campaign is a mere result of the fact that the project elicits involvement from a number of independent individuals with outstanding beliefs. Run into also (Muchnik et al., 2022) for an earlier observation congenial to the nowadays one.

In whatever case, our data do non indicate any significant influence of the previous pledges on the current pledges, as it occurs in wealth evolution (Solomon and Richmond, 2001) or citations dynamics, which are governed by the multiplicative or self-exciting processes (Golosovsky and Solomon, 2022). This ascertainment limits the predictive power of the time-pattern plumbing equipment techniques used in the by (Etter et al., 2022; Chung and Lee, 2022; Chen et al., 2022; Rao et al., 2022) for prediction success of the Kickstarter campaigns. By contrast, the mere detection of the "backers of type Two" in daily pledge distributions is an efficient and very early predictor of success, as we explicate below.

Indeed, consider the probability density functions (PDF) of the accumulated pledge distribution (Fig. 3). The PDF for failed campaigns, ρ _f (q), has most of its weight at minor daily pledge q and drops exponentially at high q-southward. This contrasts the PDFs for successful campaigns, ρ _southward (q), that exhibit a maximum at certain q _max that depends on fourth dimension. Most of the weight of the PDF for successful campaigns is located around this maximum and the weight at low pledges is strongly macerated. Thus, the PDFs of the failed and successful campaigns are distinctly different and occupy different portions of the ρ-q diagram (Fig. iii) and this prompts us to innovate a threshold. Should the two distributions accept a large overlap, one could not find a threshold separating finer the sets of successful and failed campaigns.

Conversely, if the PDFs are disjointed one could observe a threshold (in fact more one) that effectively separates the distributions for successful and failed campaigns. The actual situation is close to the second possibility: ane sees in Fig. 3 that the blue (successful) and red (failed) PDFs intersect. This allows one to cull the threshold indicated by a vertical arrow. Past plotting the position of the threshold for everyday of the entrada 1 obtains the red threshold line in Fig. 1. This threshold is a basis for our predictor of success.

Groundwork information on Kickstarter

A typical crowdfunding entrada is initiated by an entrepreneur proposing his project through the Internet platform. The entrepreneur indicates two most important parameters: the target investment necessary for performing the project and the deadline for reaching this target. Once the project is published on the Internet, any individual tin can become a backer by pledging funds to support the entrada. The number of backers and the amount of daily pledges are publicly available. If the borderline comes and the total amount of pledges equals or exceeds the target investment, and so the campaign is considered as a success and its implementation begins. If the target investment is not achieved, the campaign is considered a failure and is discontinued.

The impact of accurately predicting whether the campaign succeeds (or to what extent information technology succeeds) serves many purposes. Showtime, among the multitude of similar campaigns, such as designer shoes or computer games of a specific genre, early classification may help promoters straight their energies towards campaigns that need it most. From the perspective of potential backers, it is also of interest to evaluate the stakes at mitt when pledging: the consequences of pledging for a campaign that is likely to succeed even without their pledge, or for a ane that is more than likely to neglect, are unlike.

It is thus natural to consider predictor algorithms that care for crowdfunding equally a dynamical stochastic or causal process whose behavior can be studied, understood, predicted, and influenced. Our goal is to develop such algorithm. In contrast to motorcar-learning algorithms, we base of operations our predictor on the deep understanding of the dynamics of crowdfunding. The present study is important because it tin can be extended to similar bug arising in very different contexts such as elections, betting, economic and financial markets, etc. Recall in this context the self-referential dynamical aspects of the famous Keynesian beauty contest game (Keynes, 2022).

Previous studies of crowdfunding are summarized in (Kuppuswamy and Bayus, 2022). Chung and Lee (2015), suggested a set of predictors of success based on pledge time-serial, tweets, and entrada/capitalist graphs. These predictors achieve spectacular 76% accurateness after 4 h of campaign launch. In add-on, Etter et al., 2022, constructed an Net site which can run such predictor in real fourth dimension. Greenberg et al. (2013) employed and trained back up vector machine (SVM), and thus created a classifier organization for predicting campaign success with a 68% accurateness based solely on initial weather condition. Qiu (2013) showed that posting the campaign on the Kickstarter homepage has a positive effect on receiving pledges. Mitra and Gilbert (2014) analyzed the contents of entrada web-pages and showed that their linguistic communication may be used to meliorate prediction. Chung and Lee (2015) developed models which predict success and total amount of expected pledged money. The higher up models operate with time-resolved Kickstarter and Twitter data and yield approximately 90% accuracy when the campaign achieves thirty% stage (the time window between the launch of the project and its deadline). Chen et al. (2013) created and trained an SVM to predict the probability of success. This predictor achieves 67% accuracy at the launch phase, and approximately 90% accuracy when the campaign proceeds to the 40% phase. Rao et al. (2014) used decision-tree models to investigate the extent to which simple inflows and get-go-lodge derivatives can predict entrada success. Basing on the initial 15% of coin inflows they could predict success with 84% accuracy.

Data analysis

We analyzed the publicly available data on the Kickstarter campaigns using the database reported in (Etter et al., 2022). The fourth dimension-series data covering N ₀ = 16,043 campaigns were assessed on a daily basis (daily pledges, daily number of backers), without access to private pledges or backer specifications. Commencement, we analyzed campaigns with t ₀ = 30 twenty-four hour period elapsing as the in-sample database to determine the parameters that discern campaign success from failure. These included N _s = 3177 successful and N _f = 3964 failed campaigns, whereas the rest (4562 successful and 4339 failed campaigns) were used as the out-of-sample database on which we tested our success predictor.

We announce by Q(t) the accumulated pledge by day t of the entrada and by Q ₀ we announce the goal, namely, the target pledge. Since the projects differ greatly in their target pledge, to put all campaigns on the aforementioned continuing we considered the reduced pledge, q(t) =Q(t)/Q ₀. We constitute that the dynamics of pledge aggregating depends more on q(t) rather than on the target pledge Q ₀, hence this reduction is justified.

We divided all campaigns onto successful and failed ones and studied their statistics separately. To this end nosotros built statistical distribution of the accumulated pledges for every stage (day) of the campaign and characterized them by CDF, ${\Pi}\left( {q,t} \correct) = {\int}_q^\infty {\rho \left( {q\prime ,t} \right)dq\prime }$, where ρ is the PDF. The following relation holds $\rho _0N_0 + \rho _sN_s + \rho _fN_f$ where ρ _s and ρ _f are the PDFs for successful and failed projects correspondingly, and ρ ₀ is the PDF for all projects together.

Though we simply have data on the total daily pledges rather than individual backer pledges, we can use a large daily pledge point every bit a proxy for a large private backer pledge. Indeed, our analysis of the number of backers that contributed in each day revealed that the days with large total pledges did non brandish significantly more backers than ordinary days.

From Figs. two and 3 one sees that the set of successful campaigns has from the very get-go a signature that is quite like shooting fish in a barrel to resolve: this is the presence of fatty tails in the pledge distribution of successful campaigns which is indicative of type II backers.

In Fig. 4 one sees that there exists a line (colored in crimson in the figure) such that the vast majority of successful campaigns lie in a higher place it while the failed campaigns prevarication mostly below it. Indeed, one sees that the ruddy line is placed:

-In Fig. 4a it is to a higher place the white region which contains just the 10% successful campaigns with the lowest pledges.
-In Fig. 4b it is below the white region which contains only the ten% failed campaigns with the highest pledges.

On this basis one can predict success quite precisely on each day by establishing whether the accumulated pledges at that day are above or below the threshold.

In what follows nosotros explain how we chose the threshold q ₀ (reddish curve in Figs. 1 and 4) and why it provides a useful criterion for predicting the result of a campaign. Following Fig. 3, we consider the following function

$$F\left( {q_0} \right) = N_f \cdot {\Pi}_f\left( {q_0,t} \right) + N_s\left[ {1 - {\Pi}_s\left( {q_0,t} \correct)} \right]$$

(one)

The first term represents the total number of failed campaigns that by time t garnered full pledge exceeding q ₀, while the second term represents the number of successful campaigns that by time t garnered total pledge below q ₀. If we consider q ₀(t) as a predictor of success after phase t, the first term in Eq. (1) is the number of false positive events and the second term represents the number of false negative events. Our prediction algorithm minimizes the number of false events at each time t, namely, for each phase of the campaign we find q ₀(t) that minimizes F(q ₀).

To measure the success of our prediction algorithm we ascertain the prediction accuracy as:

$$A\left( {q_0,t} \right) = \frac{{N_s \cdot {\Pi}_s\left( {q_0,t} \right) + N_f\left[ {ane - {\Pi}_f\left( {q_0,t} \right)} \correct]}}{{N_0}}$$

(2)

Namely, for each stage of the entrada we add upwardly the number of fake positive predictions (the total number of failed campaigns above the threshold in Fig. 4b) and false negative predictions (the total number of successful campaigns beneath the threshold in Fig. 4a). Then, nosotros split this sum by the full number of campaigns.

To allow comparison with previous studies we plot on Fig. five the prediction accurateness A(q ₀ , t) reported in each of these studies as a function of the campaign phase. We find that the accurateness of our predictor exceeds that of other groups, especially in the crucial days at the beginning of the campaigns.

The threshold is a binary predictor, offering the best judge of whether a given projection is more likely to succeed or fail, regardless of how far higher up or beneath the threshold the projection's pledge trajectory is. In what follows we consider a more than authentic tool which calculates the probability of a project to succeed. We determine the probability of success of a campaign that accumulated the normalized pledge q at time t as follows:

$$P_{{\mathrm{success}}}\left( {q,t} \right) = \frac{{\rho _sN_s}}{{\rho _sN_s + \rho _fN_f}} = \frac{{\rho _s\left( {qt} \right)N_s}}{{\rho _0\left( {q,t} \right)N_0}}$$

(3)

Figure half dozen plots this probability for different days of entrada. In fact, Fig. half dozen can be used as a graphic tool similar to a "table" with continuum number of columns to read straight the probability of success as the equal probability line on which the 10-coordinates and y-coordinates, that correspond to the t and q of the campaign, run across.

As an case of the employ of this map, suppose that at a campaign stage corresponding to 70% of the allocated time, the entrada accumulated 30% of the target pledge. Nosotros observe that the corresponding dashed directly lines intersect at the green profile line representing success probability 0.5.

As it turns out, when the method for deriving Fig. 6 is applied to the out-of-sample prepare of campaigns, one obtains that simply 0.5% of the predictions are off by more than 10%.

Discussion

Our analysis shows a striking deviation between the funding dynamics of successful and failed Kickstarter campaigns, even at a very early campaign stage. Figures ii and 3 show that these funding patterns are fundamentally different, since failed campaigns display exponential pledge distribution, while successful campaigns brandish a heavy-tail pledge distribution.

For successful campaigns, the tail of the pledge distribution follows the ability-law dependence with the exponent close to unity (i.35), implying a spread of pledge sizes over a very broad scale.

Figure v shows that even limiting our method to a binary choice for the purpose of comparability with other works, our method offers better accuracy already in the kickoff three days of the campaign. Moreover, our method is transparent and may be easily applied by anyone (specially by reading the success probability from the Fig. vi). Possibly, these results may likewise help developing strategies to promote campaigns by intervening in the pledge procedure (Muchnik et al., 2022).

While the data let u.s. to exclude multiplicative random walk or cocky-exciting processes in the case of Kickstarter campaigns it is still unclear what is the origin of the heavy-tail pledge distribution and specially how to induce it. This is a crucial point to understand and reproduce successful campaigns: for Kickstarter campaigns the very proper noun "viral", that implies contamination between participating individuals, is put nether question by the nowadays data which display heavy-tail distributions from the very beginning of the campaign. Thus, the Kickstarter campaigns meliorate fit the "homophily" hypothesis (Muchnik et al., 2022), which postulates that the reason for individuals responding to the same stimulus is not contagion over their connections but the mere fact that they have like preferences.

The hope to understand the mechanism responsible for the success of a Kickstarter entrada is further encouraged by the fact that in contrast to the "black-box" approach of automobile learning or statistical inference, our assay exploits the observed beliefs of the relevant groups involved in the procedure.

Conclusions

We have synthetic an like shooting fish in a barrel and authentic tool for predicting success of a Kickstarter entrada, even at its early on phase. We found almost no correlation between the pledges fabricated at different days that would point the influence of early pledges on the afterward ones. Thus, our predictive tool is not based on time-series analysis. Our empirical findings favor an a priori intrinsic holding of the community of backers rather than the advice/reciprocal influence among them as the key to campaign success. It is left for time to come studies to check if this "not-viral" collective response is behind social success of other classes of processes such as YouTube views, Twitter tweets, Facebook, and blog likes.

Information availability

The datasets analyzed in this study have been retrieved from the Sidekick.com repository, http://sidekick.epfl.ch/data, collected by V. Etter, M. Grossglauser, and P. Thiran.

References

Chen, Thousand, Jones B, Kim I, Schlamp B (2013) Kickpredict: Predicting kickstarter success. Technical Study, California Constitute of Technology
Chung J, Lee K (2015) A long-term written report of a crowdfunding platform: predicting project success and fundraising amount. In Proceedings of the 26th ACM Conference on Hypertext and Social Media. ACM. pp 211–220
Etter V, Grossglauser 1000, Thiran P (2013) Launch hard or go home!: predicting the success of kickstarter campaigns. In Proceedings of the first ACM conference on Online social networks. ACM. pp 177–182
Golosovsky One thousand, Solomon S (2013) The transition towards immortality: non-linear autocatalytic growth of citations to scientific papers. J Stat Phys 151(1–2):340–354

ADS MathSciNet Article Google Scholar
Greenberg Doctor, Pardo B, Hariharan K, Gerber E (2013) Crowdfunding support tools: predicting success and failure. In CHI'13 Extended Abstracts on Homo Factors in Computing Systems. ACM. pp 1815–1820
Keynes JM (2018) The full general theory of employment, involvement, and money. Springer, Cham

Book Google Scholar
Kuppuswamy V, Bayus BL (2015) A review of crowdfunding research and findings. handbook of new product evolution research (Forthcoming). https://ssrn.com/abstract=2685739. Accessed 28 Oct 2022
Levy Chiliad, Solomon S (1996) Power laws are logarithmic Boltzmann laws. Int J Modernistic Phys C 7(04):595–601

ADS Article Google Scholar
Levy One thousand, Levy H, Solomon Southward (1994) A microscopic model of the stock marketplace: cycles, booms, and crashes. Econ Lett 45(1):103–111

Commodity Google Scholar
Muchnik L, Aral S, Taylor SJ (2013) Social influence bias: a randomized experiment. Science 341(6146):647–651

CAS ADS Article Google Scholar
Mitra T, Gilbert East (2014) The linguistic communication that gets people to give: phrases that predict success on kickstarter. In Proceedings of the 17th ACM conference on Estimator supported cooperative work and social computing. ACM. pp 49–61
Qiu C (2013) Bug in crowdfunding: theoretical and empirical investigation on Kickstarter. https://ssrn.com/abstract=2345872. Accessed 27 October 2022
Rao H, Xu A, Yang X, Fu WT (2014) Emerging dynamics in crowdfunding campaigns. In International Conference on Social Computing, Behavioral-Cultural Modeling, and Prediction. Springer, Cham, pp 333–340
Solomon S, Weisbuch Thou, de Arcangelis L, Jan N, Stauffer D (2000) Social percolation models. Phys A 277(i–two):239–247

Article Google Scholar
Solomon Southward, Richmond P (2001) Power laws of wealth, marketplace order volumes and market returns. Phys A 299(1–2):188–197

Commodity Google Scholar

Download references

Author information

Affiliations

The Racah Constitute of Physics, The Hebrew University of Jerusalem, Jerusalem, State of israel

Alex Kindler, Michael Golosovsky & Sorin Solomon

Corresponding author

Correspondence to Alex Kindler.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open up Access This article is licensed nether a Creative Eatables Attribution iv.0 International License, which permits employ, sharing, adaptation, distribution and reproduction in whatsoever medium or format, equally long as you requite advisable credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the commodity's Creative Commons license, unless indicated otherwise in a credit line to the fabric. If material is not included in the commodity's Creative Eatables license and your intended utilise is non permitted by statutory regulation or exceeds the permitted use, yous volition need to obtain permission directly from the copyright holder. To view a re-create of this license, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and Permissions

About this commodity

Verify currency and authenticity via CrossMark

Cite this article

Kindler, A., Golosovsky, M. & Solomon, S. Early prediction of the upshot of Kickstarter campaigns: is the success due to virality?. Palgrave Commun v, 49 (2019). https://doi.org/ten.1057/s41599-019-0261-6

Download commendation

Received: 23 October 2022
Accepted: 02 May 2022
Published: 14 May 2022
DOI : https://doi.org/ten.1057/s41599-019-0261-6

Source: https://www.nature.com/articles/s41599-019-0261-6

Posted by: morrisboally.blogspot.com