Eliciting preference effect of desktop 3D printers using choice-based conjoint analysis

Eliciting preference effect of desktop 3D printers using choice-based conjoint analysis

Daniela Zhelyazkova

22.06.2015

Eliciting preference effect of desktop 3D printers using choice-based

conjoint analysis

Daniela N. Zhelyazkova University of Groningen

Faculty of Economics and Business Master Thesis, MSc Marketing Intelligence

22.06.2015

Hyacinthstraat 65 9713 XB Groningen Tel.: +31647 668736

E-mail: d.zhelyazkova@student.rug.nl Student number: 2548690

First supervisor: prof. dr. Peter Verhoef

Second supervisor: dr. Felix Eggers

M ANAGEMENT S UMMARY

The present study aims to investigate consumer behavior towards the disruptive innovation of desktop 3D printers. Presently, there is enough evidence to consider this kind of technology of great importance and to appraise it as one bringing a lot of significant changes in people’s buying behavior and in overall daily routines. In order to give proper implications about consumer preferences and their willingness to pay for these ‘machines of the future’, a choice- based conjoint analysis is conducted. For the purpose of the analysis the most important characteristics of desktop 3D printers are defined, on the basis of an extensive research on the already available market. All the implications are grounded on a survey administered in the Netherlands.

It is concluded that the majority of people tends to value mostly the price of the printer itself and how colorful and vivid the printed product may be. Using a Latent class analysis approach, two segments with different overall preferences are formed. The two segments are also broadly discussed with respect to their background demographic characteristics and their innovativeness as consumers.

Key words: desktop 3D printers, disruptive innovation, conjoint analysis, preference effect,

willingness to pay, latent class analysis, segmentation

Table of Contents

1. I

... 5

2. L

B

... 8

2.1 What is 3D printing as a technology?... 8

2.2 Why studying 3D printers? ... 9

3. C

M

... 12

4. S

D

... 14

4.1 Method and measurement... 14

4.2 Conjoint settings... 15

4.2.1 Attributes and levels... 15

4.2.2 Choice design... 17

4.2.3 Data collection... 18

4.3 Estimation procedure ... 19

5. R

... 20

5.1 Data sample... 20

5.2 Descriptive statistics... 20

5.3 Model fit ... 22

5.4 Linear or Nominal? ... 23

5.5 Relative importance of attributes... 23

5.6 Willingness to pay ... 25

5.7 Segmentation ... 26

5.7.1 A priori segmentation based on familiarity with 3D printing... 26

5.7.2 Ex-post preference-based segmentation... 27

6. D

... 31

7. L

... 33

8. R

... 34

9. A

... 38

1. I NTRODUCTION

Imagine you need a small part of an appliance that you recently broke, and it is really irritating that you cannot fix your appliance without the missing piece. What if you can easily get this part at your home in thirty minutes from a machine with the size of your inkjet printer? Would not it be great? Well, this has already become a reality. Machines like these are sold worldwide and are called broadly “3D printers”.

The most common definition of a 3D printer that you can come across is: “machines that use additive manufacturing processes to print three-dimensional objects from a digital file”. These types of machines are, at present, being produced and sold to the end-users and not only consumed for manufacturing purposes anymore. The invention of the 3D printer dates back to 1984, when Charles Hull of 3D Systems Corp. created the first working machine of that type. But 3D printing in its early days was very expensive and not appropriate for the general market.

While advancing into the 21

century, the costs have radically dropped, and that makes the 3D printing technology more affordable and obtainable (redOrbit press, 2014). In 2001, the first desktop printer was produced by Solidimension, which allowed even the single customer to be able to possess such a machine. Although the earliest commercial devices were primitive and not very easy to use, we are now entering an era, in which companies are trying to design 3D printers that are effortless to use and applicable to the end-user’s daily routines and work. The statistics shows that the total market of 3D printers in 2014 has come up to 158 000 units or 1.6 billion dollars from which 63% is assumed to be the part of the desktop printers only.

The forecasts are that in the following four years the number of the annual sales will raise to 845 000 units, while in 2040 there will be one 3D printer for every person on the American market (see Figure 1).

Figure 1. Forecast for the worldwide end-user spending on 3D printers. Source: Statista.

1The data is available at the website of CCS Insight – a global technology analyst company

It seems that the phase of the innovator’s adoption has passed, and now is the stage of the early adopters, assuming the Bass diffusion model for growth of a new product (Bass, 1969). This kind of technology is already so accessible that one can order it from Amazon and receive also a design for product for free. The customer can choose even between different brands such as:

MakerBot, CubePro, FlashForge Dreamer, Formalabs etc. Still, these machines highly differentiate in what they can actually print; the materials they use; the speed; size of the objects reproduced, and in many other parameters. The website 3dhubs.com provides even a ranking of the best 3D printers available on the market rated by customers who are taking part in the 3D printing community. Though household ownership is mainly limited to individual hobbyists, the people’s awareness of the disruptive technology is becoming higher and higher. As the price is also rather affordable – between 500$ and 2200 – 3000$ for the desktop models, the tendency is that their demand will increase. Some analysts expect that 3D printers will become more attractive, because the process of rendering things is getting simplified, for example, by software for PC, tablets and smartphones (Malhotra, 2014; CCS Insight, 2014). Currently, 3D printers are using mostly blueprints of the designs, but it is considered that soon they will be able to work only with scanned images of the subjects. With respect to the present market of desktop 3D printers, these are used mainly for designing and prototyping and also for creating parts. Growth is driven mostly by the education sector (see Figure 2).

However, a lot of the customers by so far do not understand the idea behind the 3D printers and consider them as something to have fun with, without realizing the range of perspectives they give. As Arnaud Gagneux (VP Technology Transformation at CCS Insight) says: “Consumers will only buy 3D printers if they can see a clear use of them. To drive mass sales, manufacturers need to change the perception that 3D printers are simply a bit of fun and create sustainable demand beyond just an occasional need.” In order to do this, firms should investigate consumer preferences for this kind of disruptions, and also what features (such as cost, size, and materials) would be most valuable for them. Although exiting their infancy phase, there is still lack of research on customer preferences for desktop 3D printers.

Figure 2. Most popular usage of desktop 3D printers by February 2015. Source: 3D Hubs (a 3D printing community).

The purpose of this thesis is to specify a model for eliciting the preference effect of a desktop 3D printer, which on the one hand would give great practical insights to the companies selling these technologies. On the other, it will contribute academically by providing a clear approach determining the preference effect of a disruptive technology. 3D printers are considered to be a very radical and important innovation, and scientists usually compare them and their adoption to the one of personal computers and smartphones. They were also said to cause the “third industrial revolution” or even, as the president of USA Barak Obama once suggested, “to have the potential to create new waves for job growth” (Malhotra, 2014). Hence, the matter of the three- dimensional printing is beyond doubt of high importance.

Three-dimensional printing has existed for many years in manufacturing and is nowadays used in all kind of industries and spheres. It is not so novel anymore. This is the reason why the focus of the research will be only on the desktop 3D printers that people would buy for their households or privately owned enterprises.

As an overview, the paper is trying to provide an extensive look at the disruptiveness of desktop 3D printers and their relevance as an innovation, and determine the future tendencies in consumer behavior concerning the new technology. Namely, the thesis will establish which attributes of 3D printers will matter most for their future adoption, what would be the “ideal” 3D printer for the customers, and at what price. Further, it is attempted to identify different segments amongst the respondents, based on their preferences. In order to measure the preference effect, a choice-based conjoint analysis is conducted.

The structure of the paper is organized in a consecutive way. Firstly, a literature background is

provided, affirming the disruptiveness and end-users’ 3D printers significance. Likewise,

insights are given on the methodology of conjoint analysis, and on why it would be the best

method for conducting this research. Next, a conceptual framework is provided, in which the

different variables used in the study are described. Then, a conceptual model is built, followed

by a methodology part, which includes the data collection specifications and the formation of

variables as well as reasonable explanation for choosing them. The results of the data analysis

are provided and a discussion session included for an overview. Finally, the limitations of the

study are described.

2. L ^ITERATURE B ^ACKGROUND

2.1 What is 3D printing as a technology?

Described concisely and vaguely, 3D printing is a technology that lets you take a digital file and turn it into a physical product. 3D printers, in their basis, all rely on additive processes. In the additive process, an object is created by laying down successive layers of material until the entire object is created. Each of the layers may be observed as a thinly sliced horizontal cross- section of the eventual object. Before the actual printing takes place, a virtual design of the object one wants to create is needed. With the use of a 3D computer graphics software (such as SketchUp, FreeCAD, ShapeShifter, Rhino and many more) a CAD (Computer Aided Design) file is made. Alternatively, a 3D scanner can be used, which makes a digital copy of an existing object and puts it in a file. This way of modelling really simplifies the whole process of 3D printing and is very attractive for the customer as it can be implemented in other devices such as smartphones, but the technology is still not that well established and engineers are working on it. Besides, currently there is a lot of simplified and commercialized software for creating printable 3D designs that allows every consumer with basic computer knowledge to be able to create his/her desired object. A lot of websites and 3D printing communities divide the available programs into more professional and commercial ones

. Even there are some online tools for modelling like simscale.com, which allows creating without buying and installing special software.

In order to prepare the digital design, made via the modelling program for printing, the software of the printer first slices it into hundreds or thousands of horizontal layers. The 3D printer reads every slice and proceeds with creating the object, blending each layer together with no visible sign of the layering. There are mainly three types of methods and technologies implemented in the 3D printers so far, or, said otherwise, three different ways of combining and solidifying the materials.

The first of them is called Selective laser sintering (SLS). Described in the simplest way, this technology uses a high power laser to fuse small particles of plastic, metal, ceramic or glass powders into a mass that has the desired three dimensional shape (3dprinting.com). A laser

“draws” cross-sections of the printed part on the surface of the raw material. The advantages of this method are that it can use wide range of materials and the precision of creation and durability of the products is better compared to other methods. However, the machines available on the market that are using this technology are rather expensive and not really “desktop”. Also, the objects created are a bit porous, and in order that they have a nice finished look a special treatment is required. Next is the Fused deposition modeling (FDM). This one does not use a laser, but a nozzle, which turns on and off the flow of the material. The nozzle is heated to melt the material and can be moved in both horizontal and vertical directions by a numerically- controlled mechanism. After the layers are formed, the material hardens immediately. This method is used for more complex objects and provides better precision and strength.

Additionally, it is a lot cheaper and office-friendlier than the SLS. The shortcomings are that it is very slow and the surface of the final product turns again very rough. The third method and the most appropriate for commercial use is called Stereolithography (SLA). Its technology is a lot faster than the other methods, and the production is indeed very low cost. In the process, a laser

2Seewww.makerbot.com,www.3ders.org, and3dprintingindustry.com.

is used to trace out and cure the layers on the surface of a pool of a liquid resin. The materials that can be used here vary a lot and the machines employing the technology differentiate broadly under these criteria. Usually, the products are a lot weaker but overall the stability and durability depends on the material used. Another nice characteristic with regard to this method is that the surface of the objects looks smooth and finished.

Such an extensive review of the technology behind 3D printers is necessitated due to the complexity of the products. Technological background information will ensure that the reader will better understand the further specific and researched features of the machines and their levels.

2.2 Why studying 3D printers?

There are several existing trends that may be assumed as prerequisites for the success of the desktop 3D printers. One of them is the mass customization. In literature, there can be found a lot of evidence that increasingly people want to be treated as individual customers, with products tailored to their specific needs (Griffin & Hauser, 1993; Lamberti, 2013). Mass customization relates to the ability to provide customized products or services through flexible processes in high volumes and at reasonably low costs (da Silveira et al., 2001). The concept of mass customization emerged in the late 80s. The main idea behind it is that manufacturing is becoming customer-involved, and product production is situated around the individual needs of every customer. Mass customization systems are aligned to deliver products and services that best translate the actual choices of individual customers, and not only providing wide varieties (da Silveira et al., 2001; Salvador et al., 2009).

A lot of famous brands have already implemented the concept of mass customization in their business strategy. Like Dell for example, which assemblies personal computers with specifications set by the customers. Nike, on the other hand, recently introduced a personalization section on their website called NikeID. There, every customer can make his own runners with customized materials, forms, and colors, and receive the product at his home within four weeks. Levi Strauss also tried to provide their customers with an option to personalize their jeans. These are just a few examples that demonstrate how big companies handle mass customization and meet individual needs. Mass customization is said to be the new frontier in business competition (Pine, 1999). It adds value through the variety it that offers to the customers.

3D printers are already used in manufacturing in order to ease the process of mass customization and to reduce the costs and efficiency. This technology really lessens the production process as it only needs the blueprint of the individual’s design and it practically does everything else by itself, without assembling and without separating different parts of production (Hu, 2013). On the other hand, in some spheres there is a big contrast between mass customization and 3D printing, where the former relies on pre-assembling. Still, they both lead to no unsold finished goods inventory (Berman, 2012). As the process of mass customization is welcomed and valued by the customers worldwide, one may claim that it is an antecedent of the adoption of 3D printers. 3D printing could allow customers not only to tailor the products to their needs, but to produce them at home by themselves in no time.

Next to mass customization we should mention also the co-creation activities, because they are

becoming a common practice as well (Verhoef et al., 2010). Co-creation is an activity in which

customers actively contribute and select various elements of a new product offering. Companies do the co-creation activities in order to develop new products and services that better meet consumers’ wants and needs, and in order to decrease the high failure rates of new product introductions, especially prevalent in the consumer goods sectors (Füller, 2010). By means of their contribution in terms of giving opinions, desires, and needs, as well as applying creativity and problem-solving skills, customers take on the role of co-creators in new product development (Namibisan & Baron, 2007). There is already a lot of research on the topic what drives customers to engage in a new product development activities (Hoyer et al., 2010; Füller, 2010). The specific reasons vary from social and financial ones to self-esteem and price. 3D printers will make the inclusion of co-creation inevitable, meaning that every product, that is to be printed, may be co-created by the consumer through a CAD software. As it is observed that there is higher and higher willingness of customers to participate in co-creation activities, it is essential that 3D printing will be adopted with the same power as co-creation and in a way that it allows for it in an easier manner.

The majority of authors and scientists give a great promise of 3D printing as a technology (Berman, 2012; Campbell et al., 2011; Cronin, 2012; Harouni, 2011; Flanders, 2011). There could be found a lot of writings that also claim how disruptive the 3D printing technology is (Desjardin, 2014). The technology has even been compared to the disruptiveness of personal computers and smartphones. It is hard to predict revolutions in technologies. For instance, many people in the past did not believe in the success of the computers. Similarly, today we have enough evidence that 3D printers are bringing a lot of changes around us and they have the capacity to change even more. They will not only transform different markets but also the way we purchase products. In particular, they may change the way we think about luxurious products; change the accessibility to goods (Flanders, 2011, TED Talks), and lower the prices for different kind of products. Also 3D printing will offer more sustainable products with less waste material. As the technology of 3D printing is becoming now cheaper, it will increase the hazard of disruption (Sood & Tellis, 2010).

However, dealing with disruptive technologies often leads to companies lack systematic customer integration in the product development process. Disruptive innovations break with existing technologies and are, therefore, often beyond the scope of customer’s imagination due to its complexity, level of novelty, and wide variation of standards and quality (Dawson, 2014;

Backhaus et al., 2014). The effect may be confusion in customers, while the predicted adoption behavior may not have the according pace.

Therefore, before entering the market, companies should investigate consumer preferences as well as the needs for these innovational technologies or for the motivations behind them.

Evaluating highly complex innovations helps generating a clear picture of consumer preferences.

This has already been a practice in B2B markets with the help of virtual prototypes (Backhaus et al., 2014). The most appropriate method known so far in the literature with regard to studying consumer preferences for products is the conjoint analysis (Gustafsson, 2007; Pullman et al., 2001). In this kind of investigation, products are considered as a bundle of different attributes.

Accordingly the preferences for attribute levels are decomposed statistically from the overall

evaluation of the product. In this paper, the main aim of the conjoint analysis will be to facilitate

customers’ imagination about disruptive innovations and, practically, to give a clear picture

which of the attributes will be most important during the process of adoption as well as which

levels will fit most the different segments. Assuming that 3D printers will make a real revolution,

it is of utmost significance to reveal the consumers’ preferences for them.

3. C ^ONCEPTUAL M ^ODEL

In order to estimate the preference effect of desktop 3D printers, first a model should be specified. In the model formation, the independent variables to be included are the different attributes, which represent the most important features of the abovementioned technology, and the dependent variable is the overall preferences of a respondent for a desktop 3D printer. By overall preference, it is meant the importance that each attribute and its corresponding levels have for the respondents. During the estimation procedure, the different choices, which the respondents make, will be taken as a DV. It is processed in such a way in order to estimate preference utilities of respondents formed by the different choices. It is assumed that the respondents choose an offer that maximizes their utility (Felix Eggers & Fabian Eggers, 2011).

Based on a research over the available market of three dimensional desktop printers, the most important factors regarding their adoption are specified as follows: print speed, cost of the hardware, feature detail resolution, accuracy, material properties, colors, and ease of use. Each of them will be extensively described in the study design part, and arguments for their inclusion will be provided. The different attributes are not assessed separately but in interaction with their specified levels, and, hence, their relative importance will be measured.

Moreover, some control variables, consisting of demographic characteristics are included for segmentation purposes. These are age, gender, nationality, marital status, education, occupation and employment status. Besides, a control variable is included, measuring the willingness of the respondents to adopt an innovation, and also one measuring the respondents’ previous knowledge on 3D printing. The consumer innovativeness regarding technologies is measured by a 5-point Likert scale adapted from Goldsmith & Hofacker (1991). There will be segments specified a priori, that are based on a control variable as well as segments based on the preferences.

The incremental (relative) willingness of the respondents to pay will also be measured on the basis of the preference effect. This measurement reveals to a larger extent the price sensitivity of the respondents, or, in other words, how much more they are willing to spend for the different levels of a particular attribute. The absolute willingness to pay will be measured as well.

The finalized conceptual model, based on the most relevant attributes specified, is shown in

Figure 3.

Figure 3. Conceptual model summarizing the general research idea

Print speed

Cost of the hardware

Feature detail resolution and

accuracy

Consumer preferences

Willingness to pay

Segments

Material properties Ease of use

Colors

4. S ^TUDY D ^ESIGN

4.1 Method and measurement

The presence of a market for desktop three-dimensional printers is now beyond dispute and this triggers the need for a reliable forecasting model that predicts if the current trend is sustainable.

Moreover, this model should be highly applicable to radical innovations. There is still lack of historical data for the product concerned, so a forecasting model is required, which captures the tendencies without having the need to specify a functional form in advance (Felix Eggers &

Fabian Eggers, 2011). All these criteria are covered by the conjoint analysis. So far it has been widely used to determine the preference effect of innovational products and services and to predict the adoption behavior (Gustafsson et al., 2000; Felix Eggers & Fabian Eggers, 2011;

Wlömert & Eggers, 2014).

In terms of overall evaluation, there are different conjoint methods. Basically, the most common ones are rated-based, ranking-based, and choice-based. Compared to choice-based conjoint (CBC) analysis, the rated- and ranking-based ones have several limitations. For instance, the CBC asks for the most preferred option, as rating and ranking give sometimes unclear view what is the customer’s preference, one cannot distinguish between two ratings like 5 and 6. The choices are more like a natural manifestation and represent more actual demand patterns. In real life, customers tend to make choices rather than rate products. Besides, the model allows prediction and inclusion of a no-choice option (e.g. is the person willing to buy this product at all). The no- choice option adds-up more realism to the study. Overall, the CBC “has been shown to be superior” (Felix Eggers & Fabian Eggers, 2011) in terms of measuring a preference effect.

The preference effect comes from utilities estimates for the different attributes and their levels.

The utility of consumer n for specific dimensions of a 3D printer (i) might be expressed with the following formula:

= + ,

where V is a systematic utility component or rational utility and ε is the error term (random component). During the experiment, consumers choose repeatedly their best choice amongst set of alternatives. As mentioned before, 3D printers are represented as combinations of attributes and their levels. While selecting their preferences, consumers attach part-worth utilities to each attribute. V is a parameter vector that links product attributes/levels to preference estimates (Felix Eggers & Fabian Eggers, 2011). The following equation represents the systematic utility:

= ∑ ,

where k is the number of attributes, x is a dummy, indicating the specific attribute level of product I and β is part - worth utility of consumer n for attribute k.

The procedure of conjoint analysis follows several steps, which should be conducted

subsequently. First, the attributes and levels, which are most relevant for the specific product,

should be specified. Then the experimental design is to be set, taking into account several

requirements. Next is the planning for the data collection. After gathering the data, an estimating

technique is specified and the model fit is assessed. The final step is the interpretation of the results.

4.2 Conjoint settings

4.2.1 Attributes and levels

The first and most important step before moving to the experimental design is identifying the product attributes with the highest impact on sales and on the market (Eggers & Sattler, 2011).

There are several ways to find relevant attributes, such as in-depth interviews, focus groups, qualitative researches, managerial needs, action studies, internal brainstorming, and so on (Simmons & Esser, 2000). In this study, due to the novelty of the products, different sources were used to define the attributes. Overall, what was considered were customer reviews, online magazine publications as well as user guides for already released desktop 3D printers. The attributes that were derived are considered to be the most relevant for the competitiveness and functionality of the products. As Simmons and Esser (2000) write in their paper: “Attributes in a conjoint design are factors or things that not only describe a product or service, but also differentiate the product or service from its competitors”. The main source of information regarding the formation of attributes is a non-commercial survey conducted in 2013 within 3D communities

. The participants were asked about the most important feature that they would like to see in a 3D printer and were given a predefined list of options to choose among. Other sources with big impact were buyer’s guides of major producers of 3D printers describing the most relevant attributes of the product.

Moreover, several websites for technical specification and comparing technologies amongst different brands were taken into consideration. They juxtapose the best sellers of 3D printers with the most relevant attributes for their functionality.

When it comes to the parameters of the attributes, it is advisable that not more than six different ones are used, or otherwise the complexity for the respondent would be higher (Eggers &

Sattler, 2011). At the same time, the study is dealing with more novel and complicated products and it is recommended more attributes to be used in order to better represent the technology.

So it is best to keep the number six. Furthermore, the attributes should be also independent, which means that levels can be combined freely with one another.

In this study design, the most important dimensions of a desktop 3D printer are determined as follows: print speed, cost of the hardware, feature detail resolution, accuracy, material properties, colors, and ease of use. The print speed in overall terms is defined as the millimeters printed by second or the time required to print an object with specific dimensions (Moilanen, 2013; 3DSystems; Reece, 2013). By cost of the hardware one should understand the price of the printer itself. It is named like that because the term is very common to be related to the price of the materials as well, which does not appear to be so relevant for the market as they hold reasonable prices for now. The feature detail resolution and accuracy correspond to the quality of the final product and how the materials are combined in it. The two attributes may be

3The data and results from the survey are available athttp://surveys.peerproduction.net/2013/09/3d-printing- survey-2013/5/.

4See for example http://www.3dsystems.com/landing/3dp-buyers-guide/resources/3D-Printer-Buyers-Guide- 2012.pdf

5Such websites are:http://3d-printers.toptenreviews.com/;http://www.makershed.com/pages/3d-printer- comparison.

merged under one item for simplifying the choice design. Usually the resolution of the printed object is measured by microns, which may confuse the consumer, even though an explanation is provided. That is why the study uses an index called Print Quality Score. The PQS captures in itself the overall quality of the product, as combining speed, resolution, and accuracy, and is represented in percentages, which will be easier for the consumer to comprehend (http://3d- printers.toptenreviews.com/). The material properties may be quite an extensive concept, but under this respective feature, the various comparison websites designate the materials that a printer is able to use. The next attribute corresponds with the issue of in how many colors the printer can reproduce a subject and how realistic the output will appear. Finally, ease of use captures in itself different features such as what operational system the printer is compatible with, whether it can print from a USB or an SD card, and whether it is compatible with every modelling software or with a scanner device. Ease of use is also indexed when represented in the study design, as every level of it is provided with an explanation of how exactly it should be comprehended. It is not advisable that different features be used as levels, as this would complicate the analysis.

Eggers & Sattler (2011) write in their paper “Preference Measurement with Conjoint Analysis”

that the levels should be a realistic representation of the marketplace and that they have to be broadly acceptable to most respondents. So upon defining the levels, an account is taken first of their range available on the market of desktop 3D printers at the moment (Orme, 2002). The ranges for all of the attributes are as follows: print speed – 10 to 300 mm/sec; cost of the hardware – 350 to 3000$; PQS – 70 to 100% (which equals approximately 20 to 400 microns resolution); material properties – from ABS, HIPS, and Nylon to PLA, wood filaments, Bendlay, and Polycarbonate; colors – from one color to full spectrum of colors; ease of use – from 1 to 5 (indexed). In order to make the levels more acceptable and comprehensive, additional explanations will be provided during the survey so that the respondents have a clear idea about the features and levels of 3D printers.

In order to define the levels several criteria should be kept in mind. First, as a rule of thumb, their number should not exceed seven. It is better to have more reliable information about fewer levels, than less precisely measured information about more levels (Eggers & Sattler, 2011). Yet, by increasing the levels, the number of parameters for the estimation is also increased, and that requires more data. Next, the levels should have unambiguous meaning and each of them shall be assumed to be mutually exclusive from the others. The best option is if there is a balance between the number of levels, i.e. whether all the attributes have the same number of levels. The unbalanced number of levels may lead to higher importance of the attributes with more levels.

All the specified attributes and levels are shown in Table 1.

Attributes Levels

Print speed 15 mm/sec 30 mm/sec 80 mm/sec 200 mm/sec 300 mm/sec Cost of the

hardware 300 $ 500 $ 900 $ 1000 $ 2000 $

Feature Detail Resolution and Accuracy (PQI score)

70% 80% 90% 95% 100%

Material

Properties More simple and fragile

materials (ABS, HIPS)

ABS, HIPS,

Nylon ABS, HIPS, Bendley, Polycarbonate

PLA All kind of materials + wood filament Colors One color Basic colors Full spectrum

of colors Wide range of

colors Life-like colors

Ease of Use 1 2 3 4 5

Table 1. Specified attributes and levels

4.2.2 Choice design

The experimental design is the combinations of attribute levels that are shown to the respondents. The choice design can be either full factorial or fractional factorial. The full one consists of all the possible level combinations. In this case we have 6 attributes with 5 levels each, which mean that the full factorial design has 5

= 15 625 combinations. This is far too much to be represented in a single survey, that is why the fractional factorial design is implemented.

Each respondent was shown only a subset of stimuli to evaluate. There are criteria for selecting choice design as well (Eggers & Sattler, 2011). The attribute levels should appear an equal number of times, because if one appears more than the others, it would be considered more important by the respondents. It is the same for each attribute level pair – it should appear an equal number of times. Moreover, there should be minimal overlap, or the alternatives in a choice set must be a lot different from one another. And a final presumption is that there should not be a dominating alternative in a choice set such as, for example, a printer for the lowest price, with the highest speed, easiness of use, PQS, full spectrum of colors and all kind of materials. The alternatives should be equally attractive to the respondent. All these issues are taken into account in the survey.

With regard to the choice sets, there are 3 alternatives represented plus a no-choice option, to make the experiment more realistic. The no-choice option was included as a separate question, not next to the alternatives. The stimuli were randomly selected, with the help of the software. It was decided that there are in total 12 choice sets to be displayed to every respondent, because according to the literature, should there be more than 15 choice sets, the respondents need to be specifically stimulated to evaluate them (Eggers & Sattler, 2011). As mentioned before, 3D printers are highly complex technology, so during the survey the respondents are provided with a short concise explanation about all attribute and levels and about how these machines work.

In order to assure that the respondents have sufficient familiarity with the product, the

explanatory text is made available during the whole choosing procedure. Figure 4 represents a

single choice set of the survey.

Figure 4. A single choice set design derived from the survey

4.2.3 Data collection

The study itself is computed by means of the online tool Preference Lab. Only respondents, who live, work or study in the Netherlands are targeted for the survey. Before distributing the final version, an initial one was tested with several respondents to show how realistic the attributes and levels are. After completing the test study the respondents were asked the following questions (Simmons & Esser, 2000):

- Do they understand all of the questions, attributes and levels?

- Did the choice sets represent the decisions they follow when choosing the product or service being examined?

- Does the output reflect the weight of importance that the respondents place on each attribute?

- Do all the attributes have at least some importance for some of the respondents?

After correcting the initial survey the tested one was distributed with the help of the social

media, study and 3D printing communities, as well as networking and snowball technique were

used. Although the snowball is part of the nonprobability sampling methods, mostly were used

probability random ones. This means that the sample units were collected by chance (Malhotra,

2010).

4.3 Estimation procedure

All the estimations were run with SPSS and Latent Gold software, as for the purpose of the analysis several calculations were done with the use of Excel.

Except the already specified Utility model was used Multinomial Logit Model to estimate the probability functions. The model is calculated through the following equation:

( | ) =

,

where i represents the alternative chosen from choice set J. For the estimation purposes also a Log-likelihood function is required, expressed with the following formula:

= ∑ ∑ ln( ( | )) .

For the segmentation purposes a Latent Class Model is used. This model assumes that

consumers belong to segments that differ in their preferences. As the optimal number of

segments is not known before the analysis, models for several numbers of segments are

estimated. Then, based on the information criteria statistics the best fit amongst them is found.

5. R ^ESULTS

5.1 Data sample

There were 167 respondents, who successfully completed the survey, out of 395, who started it.

This represents a completion rate of 42.28%. The relatively low completion rate can be explained by the fact that the survey required more time to be filled in as well as more concentration due to the explanatory text. The average time for completion is about 12 minutes, as the longest one is 30 minutes and the shortest around 10. This implicates that the respondents gave considered responses. The sample size is above the threshold of 150 respondents, which is the lowest sufficient amount for this type of choice-based conjoint analysis (Orme, 2010). Every respondent was shown 12 different choice sets with 3 alternatives each. Thus, in overall, 2004 choice sets were evaluated by the respondents.

Women prevail amongst the respondents with 51.5%. The age of the respondents ranges between 20 and 56 with a mean of 28.53. The highest percentage is in the range between 23 and 29 years. The prevailing part of the respondents (85%) has university education, either a bachelor’s or a master’s degree. 55.1% of the sample are people employed while 34.1% are students, most of them single, without a partner (for all the charts describing the sample, see Figure 5 to 10).

Although the survey took place in the Netherlands, there was a question asking for the nationality of the respondents. That question was included, because there were expectations that a lot of students, as well as immigrants, would fill in the survey. The results are that the two major groups are Bulgarians with 58.7% and Dutch with 21.6%, where the rest is formed by various other nationalities. This may be explained by the procedure of gathering the data, or by the means of communicating the survey to the respondents – it was posted in social media groups for Bulgarian ethnicities, who live in the Netherlands and for students of Bulgarian origin. They seem to have the highest response rate according to the statistics.

5.2 Descriptive statistics

Some further descriptive statistics are added in addition to the data sample description. The following are assumed to be more determining for the formation of preferences. First, the occupation of the respondents is considered to be a driving factor when it comes to acceptance of technologies (Dickerson & Gentry, 1983). The field of occupation of the respondents, who filled in the survey, is shown in Table 2 in Appendix 1. The sectors that prevail are Management, Business and Financial Operations, Computer and Mathematics, Sales and Related occupations.

Next to the occupation, respondents were asked for their previous knowledge of 3D printing technology. Surprisingly, as can be seen in Figure 9, 6.6% of the population has already used 3D printers, while 21% has a good knowledge of it, and only 5.4% has never heard of it. The prevailing part of 67.1% has only heard or read about the innovation technology. These findings suggest that next to the explanatory text and media in the survey, people were already familiar with the disruptive innovation, which increases the validity of the research.

With regard to the no-choice option, 65% of the responses were positive, or, otherwise said, the

individuals chose more times that they would actually buy a product with the same

specifications, as selected. This also increases the validity as it is assumed that the respondent got good familiarity with the product.

With respect to the willingness to adopt an innovation, the participants tend to be more in a middle position when asked to rate a statement. For the positive statements, the mean is more to agree with them, whereas for the negative the mean is more to disagree.

Figure 5. Frequencies for gender Figure 6. Frequencies for marital status

Figure 7. Frequencies for employment status Figure 8. Frequencies for education

Figure 9. Frequencies for knowledge of 3D printing 0,0%

10,0%

20,0%

30,0%

40,0%

50,0%

60,0%

Male Female

Gender

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

Marital Status

10,0%0,0%

20,0%

30,0%

40,0%

50,0%

60,0%

Student Intern Employed Not

employed Disabled, not able to

work

Employment Status

10,0%0,0%

20,0%

30,0%

40,0%

50,0%

60,0%

Education

0,0 10,0 20,0 30,0 40,0 50,0 60,0 70,0 80,0

I have never heard of 3D printing

before

I have heard/read

about 3D printing I have good knowledge of 3D printing and how it

works

I have used a 3D printer

Knowledge of 3D printing

Figure 10. Willingness to adopt an innovation

5.3 Model fit

The fit of the model is tested by comparing it to a NULL model. Or expressed otherwise, the NULL model sort of predicts rather randomly what respondents are choosing, or β = 0. This is also called a Likelihood Ratio Test. When conducting it, we hypothesize that there is no difference between the two models. The Log-likelihood for the NULL-model is calculated with the corresponding formula:

(0) = ∗ ∗ ln 1

The parameters are shown in Table 3. The Log-likelihood of the aggregate model LL(β

) equals - 1878.42, and the LL(0) results in -2201,62. Hence, from the results one can assume that the specified model is better than the NULL-model, since LL(β

) >LL(0). This is not the final presumption, as the R

and Chi-Square statistics have to be taken into account as well. All statistics are shown in Table 4 and Table 5. Overall, the prediction model is good, as the hit rate is about 55%, compared to the one of the NULL-model, which predicts only 33%. It is considered that a hit rate above 55% indicates a good degree of predictive validity (Papies et al., 2011).

Table 4. Goodness of fit compared to the NULL model

,0 2,0 4,0 6,0 8,0 10,0 12,0 14,0 16,0

Unwilling Unwilling Neutral Willing Willing

Innovativeness

Innovativeness

LL(β) -1878,4198

L(0) -2201,619026

Pseudo-R²

R²=1-(LL(β*)/LL(0)) 15%

R²adj=1-((LL(β*)-β)/LL(0)) 14%

Likelihood Ratio Test

H0 no difference between models

Chisq=-2*(LL(0)-LL(β*)) 646,398453

P-Value <0.00001

Hit Rate 55%

Number of respondents (n) 167 Number of alternatives (m) 3 Number of choice sets (c) 12 Number of parameters (β) 25

Table 5. Parameters of the NULL model

5.4 Linear or Nominal?

Since there is no general intercept as, for instance, in the linear regression, the model specification for each attribute may vary between linear (numeric), quadratic, and nominal (part-worth). When deciding about the form for an attribute, it is not necessary that it is the same for all of the attributes. For most of the attributes used in the model, it makes sense to use part-worth modelling, because it assumes no specific relationship, and also because separate parameters for each of the attribute levels have to be estimated. The price is more realistic to have a linear function. Thus, a comparison will be made between a model with a price attribute taken as numeric once, and, secondly, taken as nominal. All other attributes will be taken as nominal. Based on Table 6, it seems better to estimate the price as a nominal variable, as the adjusted R

of this model is higher than the one with linear price variable. Moreover, if one plots the utility estimates of price, its functional form might be seen to a greater extent as nominal (see Figure 11). Hence, the consecutive analysis will be conducted with all the variables taken as nominal.

Figure 11. Functional form of the price utilities

LL npar Hit rate R² R²adj

LL price part-worth -1878,4198 25 55% 0,14680071 0,13544543

LL price linear -1893,083 22 54% 0,14014052 0,13014787

Table 6. Price linear versus nominal model

5.5 Relative importance of attributes

In order to measure how much influence each attribute has on people’s choices, the parameters output for 1-class model is taken. For each attribute, the range between levels is calculated, or, in other words, the smallest β is extracted from the highest one. Afterwards, the range per each attribute is divided by the total and it results in the relative importance per attribute. All the observations are shown in Table 7.

300 $ 500 $ 900 $ 1000 $ 2000 $

Utility 0,3355 0,2936 0,0817 -0,0275 -0,6833 -0,8

-0,6 -0,4 -0,2 0 0,2 0,4

Price functional form

Attributes β Range Importance Print speed

15 mm/sec -0,2994 0,6058 13%

30 mm/sec -0,1946

80 mm/sec 0,0114

200 mm/sec 0,1762

300 mm/sec 0,3064

Cost of the hardware

300 $ 0,3355 1,0188 22%

500 $ 0,2936

900 $ 0,0817

1000 $ -0,0275

2000 $ -0,6833

Resolution and Accuracy PQScore

70% -0,4759 0,8694 19%

80% -0,2067

90% 0,07

95% 0,2191

100% 0,3935

Materials

More simple and fragile materials (ABS, HIPS) -0,2542 0,6653 14%

ABS, HIPS, Nylon -0,0456 ABS, HIPS, Bendley, Polycarbonate 0,088

PLA -0,1992

All kind of materials + wood filament 0,4111

Colors

One color -0,6055 0,9275 20%

Basic colors -0,283 Full spectrum of colors 0,2699 Wide range of colors 0,2965 Life-like colors 0,322 Easeofuse

1 -0,3559 0,536 12%

2 -0,0971

3 0,1118

4 0,1611

5 0,1801

Total 4,6228

Table 7. Overall relative importance of the attributes

It can be concluded that the price is the most important attribute of a desktop 3D printer, followed by the colors. The one of least importance seems to be the ease of use with mere 12%.

The print speed and the materials imply to be also of less priority to the respondents. The resolution and accuracy score is relatively high in the ranking with 19%.

5.6 Willingness to pay

In order to investigate the price sensitivity of customers, a desktop 3D printer with average characteristics is taken (80 mm/sec print speed; 90% PQS; materials: ABS, HIPS, Bendley, Polycarbonate; wide range of colors; ease of use 3). If a single consumer has to consider buying this product, or to decide not buying anything, every price below 2000$ will increase the probability. This is so, because at a price of about 2000$ the chance of choosing none option or buying is 50%. Conversely, every price above 2000$ will increase the probability of not buying.

For a better visual understanding, look at Figure 12.

Figure 12. Willingness to pay compared to the no-choice option

3D printer 3D printer

80 mm/sec 0,0114 80 mm/sec 0,0114

900 $ 0,0817 500 $ 0,2936

90% 0,07 90% 0,07

ABS, HIPS, Bendley, Polycarbonate 0,088 ABS, HIPS, Bendley, Polycarbonate 0,088 Wide range of colors 0,2965 Wide range of colors 0,2965

3 0,1118 3 0,1118

Utility sum 0,6594 Utility sum 0,8713

no_choice 0,0119 no_choice 0,0119

Probability 65% Probability 70%

Table 8. Price sensitivity

Table 8 depicts how the probability changes by lowering the price for the same type of product.

0%

10%

20%

30%

40%

50%

60%

70%

80%

2000$ 1000$ 900$ 500$ 300$

Probability

Probability of choosing no-choice

5.7 Segmentation

5.7.1 A priori segmentation based on familiarity with 3D printing

When using latent class analysis for segmentation purposes, the number of segments is not known beforehand. The latent class procedure clusters the respondents based on the similarity in their choice behavior. Before conducting a latent class analysis, an a priori segmentation was performed based on the covariate of previous knowledge of the respondents. The aim is to distinguish whether there are differences between, on the one hand, people, who already got in touch with the new technology, and, on the other hand, those who were not familiar, or only barely familiar with 3D printing. The segmentation was executed by forming two segments:

respondents, who already used a 3D printer and are highly familiar with the technology; and respondents, who never heard of 3D printing or have low familiarity. Their proportion is 121 to 46.

The p-value statistic indicates that all the estimated preferences are not significant, as they are above the threshold of .05. This means that the difference between the segments is not significant. A possible explanation for these results may be the matter that the survey aimed at making the respondents well familiar with the technology, before measuring their preferences.

Their overall preference differentiates only in terms of the attributes colors and ease of use.

Although they have almost the same preferences, they put different relative importance to the attributes. With regard to the segment with low familiarity, it turns out that the colors are most important, followed by the price, while the least importance is attached to the printing speed.

For those people who are well familiar with the technology, the importance of the price has a leading role indeed, whereas the least important is considered the ease of use. All the estimates are shown in Table 9.

Attributes Not or low

familiarity Highly familiar

or used Wald p-value Wald(=) p-value Relative importance Segment 1 Segment 2 Print speed

15 mm/sec -0,2129 -0,5625 75,2443 4,40E-13 8,1583 0,09 11,33% 15,63%

30 mm/sec -0,197 -0,1605

80 mm/sec -0,0451 0,1762

200 mm/sec 0,1657 0,2008

300 mm/sec 0,2894 0,3461

Cost of the hardware

300 $ 0,317 0,4304 148,682 3,70E-28 3,5973 0,46 21,53% 22,35%

500 $ 0,2899 0,2839

900 $ 0,0427 0,1888

1000 $ -0,0122 -0,0344

2000 $ -0,6373 -0,8687

Resolution and Accuracy (PQScore)

70% -0,4508 -0,6034 127,991 7,40E-24 8,7138 0,07 18,15% 19,67%

80% -0,1223 -0,4189

90% 0,0355 0,1932

95% 0,1839 0,2889

100% 0,3537 0,5402

Materials More simple and fragile materials (ABS,

HIPS) -0,1796 -0,4842 88,3959 9,80E-16 7,2809 0,12 14,43% 16,98%

ABS, HIPS, Nylon -0,034 -0,0551 ABS, HIPS, Bendley,

Polycarbonate 0,0885 0,088

PLA -0,2573 -0,0515

All kind of materials +

wood filament 0,3824 0,5028

Colors

One color -0,6238 -0,5935 173,713 2,10E-33 1,307 0,86 21,58% 16,10%

Basic colors -0,3067 -0,226

Full spectrum of colors 0,2754 0,2727 Wide range of colors 0,3328 0,2045 Life-like colors 0,3222 0,3423 Ease of use

1 -0,3572 -0,3496 55,4327 3,60E-09 3,2851 0,51 12,97% 9,27%

2 -0,1483 0,0323

3 0,1377 0,0424

4 0,1501 0,1891

5 0,2178 0,0858

Table 9. A priori segmentation estimates

5.7.2 Ex-post preference-based segmentation

The a priori segmentation defines the segments on the basis of socio-demographic information, and it is used in this study mostly for explorative reasons. The more relevant classification of consumers should be made on the basis of their preferences. For these reasons, a Latent Class model was implied with the assumption that the consumers belong to segments that differ in their preferences. The optimal number of segments is not known prior to the analysis and several solutions have to be estimated in order to find the best number of segments. The Latent Class procedure is probabilistic in its essence. A single consumer belongs to each segment with a certain probability. The parameters of the Latent Class are estimated by the maximum likelihood (Vermunt & Magidson, 2002).

Before defining the appropriate number of segments, first it has to be decided whether

covariates should be included to improve the classification. In order to make this decision, two

analyses were run, with and without the covariates, for classes between 2 and 8 solutions. The

covariates should help decreasing the error term and, if it appears so, it is better to keep them in

the formation of segments. As it turns out in this case, the overall classification is improved

through adding the covariates (see Tables 10 and 11). The biggest issue when implying Latent

Class Analysis is deciding about the number of segments. In order to do so, several information

criteria statistics should be considered. In this case, the smallest BIC and CAIC were examined,

because they have higher penalties for. As it is observable in Table 10, there is only one solution

for the clustering with covariates included, and it implies that the present data should be split

into two segments. After the two segment solution, when increasing the number of parameters,

the degrees of freedom become negative.

Number of

segments Log likelihood BIC(LL) AIC(LL) CAIC(LL) Npar L² df p-value Class.Err. R²(0)

2 -1818,7622 3898,542 3739,5243 3949,542 51 3637,5243 116 2,2e-681 0,0513 0,2237 3 -1765,0382 3924,162 3684,0764 4001,162 77 3530,0764 9,00E+01 7,9e-679 0,0773 0,2734 4 -1715,5342 3958,2217 3637,0683 4061,2217 103 3431,0683 6,40E+01 2,1e-679 0,0739 0,3296 5 -1673,9699 4008,1611 3605,9399 4137,1611 129 3347,9399 3,80E+01 1,7e-685 0,0723 0,3688 6 -1645,0684 4083,4259 3600,1369 4238,4259 155 3290,1369 1,20E+01 3,6e-701 0,0767 0,4003 7 -1607,4565 4141,2698 3576,9129 4322,2698 181 3214,9129 -1,40E+01 . 0,0634 0,4235 8 -1583,7694 4226,9635 3581,5388 4433,9635 207 3167,5388 -40 . 0,0508 0,4458

Table 10. Latent class segmentation without covariates

Number of

segments Log likelihood BIC(LL) AIC(LL) CAIC(LL) Npar L² df p-value Class.Err. R²(0)

2 -1730,5856 4269,8143 3777,1713 4427,8143 158 3461,1713 9 4,8e-742 0,0009 0,2222 3 -1625,8286 4740,9934 3833,6572 5031,9934 291 3251,6572 -124 . 0,0013 0,2709 4 -1567,6273 5305,2839 3983,2545 5729,2839 424 3135,2545 -257 . 0,0011 0,2995 5 -1493,5466 5837,8157 4101,0932 6394,8157 557 2987,0932 -390 . 0,0005 0,3391 6 -1448,0044 6427,4245 4276,0087 7117,4245 690 2896,0087 -523 . 0,0003 0,3587 7 -1389,557 6991,2229 4425,1139 7814,2229 823 2779,1139 -6,56E+02 . 0,0003 0,3799 8 -1336,0368 7564,8758 4584,0737 8520,8758 956 2672,0737 -7,89E+02 . 0,0002 0,4074

Table 11. Latent class segmentation with covariates included

The picked model splits the respondents into two segments with 85 and 82 cases, respectively, included in each of them. The next step in the investigation is to identify what kind of attributes and characteristics have the most preferred product per segment. For this purpose, all the β’s per each level were estimated and considered (see Table 12).

Attributes Segment 1 Segment 2

Relative importance Segment 1 Segment 2

Print speed 9,8% 16,2%

15 mm/sec -0,3596 -0,2647

30 mm/sec -0,1542 -0,2344

80 mm/sec 0,0044 -0,0386

200 mm/sec 0,1451 0,2716

300 mm/sec 0,3643 0,2661

Cost of the hardware 28,8% 23,2%

300 $ 0,8724 -0,2913

500 $ 0,564 0,0588

900 $ 0,2354 -0,0122

1000 $ -0,408 0,4673

2000 $ -1,2639 -0,2227

Resolution and Accuracy PQScore 19,4% 9,2%

70% -0,7761 -0,178

80% -0,3083 -0,1449

90% 0,0824 0,0642

95% 0,3395 0,1237

100% 0,6625 0,135

Materials 10,6% 19,2%

More simple and fragile materials (ABS, HIPS) -0,3581 -0,1771

ABS, HIPS, Nylon 0,0453 -0,1653

ABS, HIPS, Bendley, Polycarbonate 0,231 -0,0326

PLA -0,3438 -0,0759

All kind of materials + wood filament 0,4256 0,451

Colors 21,5% 13,4%

One color -1,0548 -0,2337

Basic colors -0,3936 -0,1732

Full spectrum of colors 0,4064 0,2046

Wide range of colors 0,5024 0,0934

Life-like colors 0,5396 0,1088

Ease of use 9,9% 18,9%

1 -0,4289 -0,3943

2 -0,0684 -0,1225

3 0,0921 0,1915

4 0,0999 0,2232

5 0,3054 0,1021

Table 12. Estimates for 2 segments solution

Obviously, segment 1 strove for perfection and tried to maximize their overall utility with every choice. The ideal product for them is the one with the best characteristics and the lowest price.

On the other hand, segment 2 pointed out the 1000$ as a favorable price, 200 mm/sec as a satisfactory printing speed, 100% for resolution and accuracy, full spectrum of colors, and ease of use at level 4. If one takes a look at the relative importance of the attributes, it appears that segment 1 was leaded by the price, resolution, and colors, and all the attributes have quite a low importance. It may be concluded that this segment tends to choose products only with a low price and to maximize the utilities of the other characteristics. As for the second segment, the relative importance is more equally distributed, except for the resolution, which means that this characteristic matters slightly for these respondents. Moreover, this segment was not leaded with the same ratio by the price, and it seems that it considered relatively all of the features of a desktop 3D printer. This makes the segment more valuable for researchers and managerial decisions. When it comes to the no-choice question, the second segment has a higher probability to buy the product than the first one. 48% of the respondents in the second segment pointed that they would actually buy the product compared to the other one where only 39% answered positively.

Regarding the socio-demographic characteristics of the two segments, it turns out that there is

not a significant difference between them, as the p-values for all the estimates are above .05. The

age is equally distributed between the two clusters and both have mostly (84%) young people

between 20 and 30 years of age. Concerning the gender, the second segment has slightly more

male respondents (58%), whereas in the first one the women are prevailing (58%). There is no

difference in the profiles regarding education and marital status. Taken relatively, the number of

students in the first one is a bit higher than in the second segment. Most of the unemployed people (63%) are in the second cluster. Another observation is the differentiation in the degree of innovativeness of the respondents. Segment one is more unwilling to adopt innovative technologies or is neutral. On the opposite side, the second one has a prevailing number in terms of people willing and open to adopt innovations (56%). The profile characteristics of the two segments are shown in Table 13. Here, only the results from the analysis are represented, while further and more deep implications will be reviewed in the discussion part of the current paper.

Segment 1 Segment 2

Age 20-30 84% 84%

31-40 6% 4%

41-56 10% 12%

Gender Male 42% 58%

Female 58% 42%

Education No degree 3% 0%

Secondary school 3% 3%

Community college (MBO) 2% 3%

College (HBO) 8% 4%

Bachelor 33% 42%

Master 50% 46%

Ph. D. 1% 1%

Marital status Single, never married 41% 48%

In a relationship, but not living together 23% 22%

Cohabitating 21% 16%

Married 13% 12%

Divorced 1% 1%

Widowed 0% 1%

Employment Student or intern 42% 33%

status Employed 54% 57%

Not employed 4% 10%

Occupation Service and production oriented

occupations 6% 13%

Business and finance oriented

occupations 38% 42%

Health and social occupations 12% 16%

Computers, math, engineering 14% 9%

Arts, sports, media 5% 10%

Legal occupations 10% 6%

Knowledge about 3D

printing Not or low familiarity 69% 77%

Highly familiar or used 31% 23%

Innovativeness Not willing to adopt 37% 35%

Neutral 18% 9%

Willing to adopt 45% 57%

Table 13. Profile of the two segments

6. D ^ISCUSSION

Dealing with radical innovations is a tricky business. New technologies give a lot of opportunities, but depending on their novelty and complexity, it may be quite confusing for the customers. It is evident in literature that their behavior becomes more complicated, and their diffusion pattern and preferences, according to the adoption behavior, are being increasingly studied (Jun & Park, 1999; Lee et al., 2004; Felix Eggers & Fabian Eggers, 2011). There are various approaches and techniques developed so far to investigate consumer behavior when dealing with technology innovations (Reinhardt & Gurtner, 2010). The one that appears to be most eminent and used nowadays is the conjoint analysis. Depending on the situation, there are different approaches for conducting such an analysis. It is suitable for radical innovation products, because it puts the customer in a virtual reality (Backhuis et al., 2014), where he is supposed to choose between different specifications. It resembles a shopping environment, but at the same time it is easier to make the customer familiar with the product. When it is properly specified, the conjoint analysis becomes an easier measure to understand the consumer behavior towards a market, product, or a service. It is like a strategy of probe and learn (Lynn et al., 1996).

This paper aims at specifying a model that will help investigating the preferences of consumers for desktop 3D printers. Desktop 3D printers are assumed by the scientists to be a disruptive innovation that will lead to a lot of numerous changes in people’s lives; changes that are similar to the ones brough by the personal computer and the smartphones. All the available forecasts show that in the near future it is expected that the curve of the diffusion will go higher and the market will only grow bigger and bigger. What is more, most analysts and practitioners consider that this technology has the power of changing the way people buy products, which will lead to a big transformation on the markets as well as in the economy as a whole. At the present moment, the adoption of these products is still in its early stage and it is of great importance that customer’s preferences and their behavior towards using such a product be investigated. This kind of research is crucial for firms and managers, who want to be sure that the pace of the adoption and growth of the market will increase (Eggers & Sattler, 2011; Felix Eggers & Fabian Eggers, 2011).

In line with the previous thoughts, the main idea of the current paper is to make a descriptive analysis, and to find the main preferences of the customers and their price sensitivity regarding desktop 3D printers. Such a descriptive analysis provides also a lot of implications on what features should be paid more attention to by the engineers; how valuable the current state of these machines is, and which characteristics can be neglected in favor of others. Moreover, it is evident from literature that consumers generally have heterogeneous preferences (Boxall &

Adamovicz, 2002). Then there are needed models and forecasting techniques that take into account individual characteristics of the consumers. Accordingly, a segmentation analysis was conducted in this study, in order to find homogeneous groups in a heterogeneous population.

However, some researchers consider dealing with heterogeneity with a priori analysis:

“including demographic parameters in demand function directly”; and others by stratifying the

consumers in various segments, based on their preference behavior, and which number is not

known beforehand (Boxall & Adamovicz, 2002). In order to examine the difference in between,

both a priori and a latent class segmentation were conducted.