Team Water Bottle Final Analysis Report
A Report on Water Bottle Design Attributes and Their Market Implications
Abstract
This research project will attempt to identify the optimal configuration of a reusable water bottle targeted towards young adults aged 18 to 30. Reusable water bottles come in a range of sizes and configurations. Customers can choose options with different prices, capacities, materials, lid sizes, and drinking interfaces. All of these attributes play a role in a person’s decision on what reusable bottle to buy. This leads us to the following research questions:
RQ1) What combination of attribute configurations will result in the highest potential sales?
RQ2) How much utility is derived from each identified attribute?
Introduction
Reusable water bottles are eco-friendly alternatives to single-use plastic bottles. The rise in consumer awareness about environmental issues along with the materialistic facet of social media has fueled demand for these bottles in a practical and social context. Additionally, reusable water bottles come in a variety of designs to cater to consumer preferences.
According to an article published by the New York Times, reusable water bottle revenue from 2022 eclipsed two billion dollars in the USA alone(1). With the average reusable water bottle coming in at 30 dollars (2). This estimates that nearly 67 million reusable water bottles were sold as of 2022. With an annual growth rate projection of 4.3% (3), we can estimate a market capitalization of 2.37 billion dollars. Approximately 79 million reusable water bottles will be sold in 2025.
Survey Design
Eligibility Requirements
Our only hard requirement is that the respondent must be 18 in order to take the survey. Respondents are further filtered out based on screening questions and respondent survey behavior statistics, such as if they indicated that they never use a reusable water bottle.
Desired Respondent Information
Our survey has two sections that surround the conjoint choice questions. Our survey begins with a screening section. This section aims to gain information about a respondent’s use of reusable water bottles. Following that, there are the conjoint questions. After, there is a demographics section that asks for a respondent’s age and gender.
Educational Material
In order for the respondents to be properly informed on the decisions they are making, we included a written description of each of the different attributes. We also included example images for the capacity and drinking interface attributes to contextualize choices. Our survey, as seen from the respondent’s POV, can be seen in the Appendix section.
Product Attributes and Levels
Below is a list of the different attributes that were included:
| Attribute | Range/Levels | Units |
|---|---|---|
| Price | 10, 15, 20, 25, 30, 35, 40, 45, 50 | USD |
| Capacity | 24, 32, 64 | fl oz |
| Lid Style (opening size/interface size) | large/wide, large/narrow, large/straw, small/narrow | n/a |
| Insulation | Insulated or Not | n/a |
| Material | Plastic or Metal | n/a |
The idea behind the lid style attribute is based on a generalization with two other attributes: lid opening size and the drinking interface. We decided that the size of the lid can be grouped into two sizes: large and small. We decided that the drinking interface can be grouped into three methods: a straw, a narrow hole, and a wide hole. The nuance here is that two of these combinations can’t exist due to feasability and practicality; these are a small lid size with wide hole, and a small lid size with a straw.
As a result of this qualitative analysis, we determined that our conjoint questions would be better if we didn’t burden the respondent with the two attributes, lid size and drinking interface. So, we combined them into the attribute lid style, and showed four images of the most common water bottle tops that correspond to the four possible combinations of lid size and drinking interface.
Major Changes from Pilot to Final Survey
We made three key changes between our pilot and final version. First, we replaced the text-based description of the drink interface with a picture that incorporates both lid size and drinking interface, improving clarity for survey participants and simplifying our model. Second, we redesigned the conjoint questions by replacing the information-heavy buttons, which were difficult to read and visually unappealing, with a cleaner table format. Lastly, we made additional adjustments, including removing the maintainability attribute and switching the price representation to a linear scale.
Example Conjoint Question
The survey provided six different conjoint questions containing three alternatives each. We did not include a no-choice alternative because it would still be better to get the best of the worst than to not get anything at all.
Data Analysis
Sample Description
The total sample size was 321 respondents, captured from November 18 - December 7, 2024. These respondents were asked to answer six choice-based questions, each with three alternatives. Below, we provide four visualizations of responses to questions in our survey that we deemed important to characterize our sample population. Each chart also has a description of relevant summary statistics if applicable.
In descending order, 71 respondents bought 2-3 reusable water bottles in the past year, 47 bought one, 30 bought four or more, and 16 did not buy any.
98 respondents use a reusable water bottle daily, 52 a few times per week, and 14 a few times per month.
85 of our respondents were female, 75 were male, and 4 were non-binary.
Based on the histogram, we can see that the distribution of respondent’s age is right-tailed. The median age was 37, the mean was 39.2, the max age was 73, and the minimum age was 19.
Data Cleaning
Here, we describe how we elected to filter our data to account for bad respondents and our justification if needed.
We filtered respondents that indicated they were not 18 or older, did not read and understand the consent page, or selected “Never” for how often they respondent uses a reusable water bottle. We felt that those who never use a reusable water bottle would not be invested in our conjoint questions or be able to analyze the attributes appropriately. This dropped our number of respondents to 207.
Next, we filtered out all respondents who did not answer all 6 conjoint questions or answered all 6 questions with the same answer choice. This left us with 184.
Lastly, the top 10 percentile of respondents that finished the fasted were removed. This left us with a final sample size of 165.
Modeling
Baseline Utility Model
\[ u_j = \beta_1 x_j^{price} + \beta_2 x_j^{capacity} + \beta_3 \delta_j^{large/wide} + \beta_4 \delta_j^{large/narrow} + \beta_5 \delta_j^{large/straw} + \beta_6 \delta_j^{insulated} + \beta_7 \delta_j^{metal} + \varepsilon_j \]
Summary Table of Utility Coefficient Estimates and Standard Errors
| term | estimate | std.error |
|---|---|---|
| price | 0.0003886 | 0.0031292 |
| capacity | -0.0011679 | 0.0022490 |
| top_large_wide | 0.1284213 | 0.1146273 |
| top_large_narrow | 0.1164315 | 0.1123951 |
| top_large_straw | 0.0971663 | 0.1114151 |
| insulation_Insulated | 0.0745892 | 0.0783014 |
| material_Metal | 0.1361129 | 0.0779655 |
Summary Chart of Utility Coefficient Estimates and Standard Errors
Summary Chart of a Fully Dummied out Logit Model
To be clear, we do not use the model below for any of the subsequent analysis of this report. It is included so that a quick comparisons between individual levels can be seen.
Mixed Logit Model
Sub-group Analysis
The sub-group analysis shows us that there are different preferences depending on age brackets.for the sub 38 group, they were found to prefer metal insulated bottles with a large wide top. Interestingly, with those who were above 38, places a stronger emphasis on their bottle tops rather than material or insulation.
Limitations
We would like to preface our results, especially those concerning WTP, by stating that a key limitation in the survey was the inconclusive nature of the response to the price variable.
We found an almost uniform distribution of prices in chosen alternatives. The results of our modeling suggested a near zero coefficient for price. This surprised us, as it implies that price has next to no influence on respondents alternative selection.
In future research we would want to provider a larger range of prices to force some sort of data on price preference. We had a similar issue with capacity. Also a larger sample size might reduce noise.
Our coefficient for capacity was also near zero. In future analysis researchers could, again, include a larger range of values and have a larger sample size.
Model Results
Simulated Market Scenarios
Sensitivity Analysis
Final Recommendations and Conclusions
In our abstract, we explained that our project’s goal was to configure an optimal water bottle for the young adult age range, 18-30. Based on the attribute coefficients shown in our subgrouping chart for respondents aged 37 or less, we conclude that a water bottle with metal material and insulation are the most important attributes to have. Although we expected the price coefficient to be negative, it may make sense that the sample population has a wide variance for their willingness to pay for a reusable water bottle based on cultural, economic, or generational factors. Similarly with capacity, the sample population’s lifestyles probably has too high a variance to capture a single, accurate \(\beta\) with such a small sample size.
The way our conjoint questions are displayed, we show a lid style image as part of our table of attributes. Having an image among text may have unintentionally detracted respondents’ attention away from other key attributes such as price.
Futher Limitations
For the top attribute we used pictures of real-life water bottles, cropped to only show the lids. There is a possibility that respondents saw the tops, recognized the water bottle it was from, and responded with a biased answer. Top was also the only attribute to use a picture.
Our power analysis indicates that we would need a sample size of roughly 300 to get a 95% confidence interval for our coefficients other than price and capacity. After filtering respondents with our selection criteria, we ended up with 164 respondents.
Appendix
Survey from the Respondent’s POV
References
A. Gander, “Do water bottles really need to tell us how to drink?,” The New York Times, Aug. 11, 2023. [Online]. Available: https://www.nytimes.com/2023/08/11/health/water-bottles-beverages.html. [Accessed: Sep. 18, 2024].
“Prices for mid-range water bottles,” IndexBox, [Online]. Available: https://www.indexbox.io/search/average-price-of-reusable-water-bottle/#:~:text=Prices%20for%20mid%2Drange%20water,fall%20between%20%2420%20and%20%2440. [Accessed: Sep. 18, 2024].
“U.S. reusable water bottle market report,” Grand View Research, [Online]. Available: https://www.grandviewresearch.com/industry-analysis/us-reusable-water-bottle-market-report#:~:text=The%20U.S.%20reusable%20water%20bottle,the%20form%20of%20plastic%20waste. [Accessed: Sep. 18, 2024].