library(tidyverse)
## Warning: package 'dplyr' was built under R version 4.4.3
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr     1.1.4     ✔ readr     2.1.5
## ✔ forcats   1.0.0     ✔ stringr   1.5.1
## ✔ ggplot2   3.5.1     ✔ tibble    3.2.1
## ✔ lubridate 1.9.4     ✔ tidyr     1.3.1
## ✔ purrr     1.0.4     
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag()    masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
library(knitr)

Aims and Objectives

The aim of this experiment is to determine the best total fermentation time to maximise sweetness, when adding fruit juice at different temperatures to flavour kombucha.

For this research, the objectives are:

  1. Determine the effect of different total fermentation times. 

  2. Determine the effect of different initial temperatures on secondary fermentation times.

  3. Determine the effect of different fruit products (concentrate, frozen and fresh).

  4. Analyse the effects on the sweetness attribute of Kombucha.

As Kombucha contains a health halo for its nutritional benefits (probiotics, antioxidants, and acetic acid), it is important for kombucha to also taste appealing. Allowing for a wider consumer audience to be able to enjoy the taste of kombucha but also reap the health benefits. Currently, Kombucha is perceived to have a sour, tart and/or light vinegar flavour profile which may be off-putting to certain consumers.

Response Variables

Perceived sweetness of the kombucha samples was the response variable for this experiment. A mixture of Triangle and Tetrad sensory tests were used as a Hedonic measurement for the attribute. In Triangle tests panelists are required to single out one of three samples that they believe is different to the others. Similarly the Tetrad tests require panelists to group two out of four samples instead. Both tests were assessed using a continuous line scale that was converted to a 0-10 point scale. The use of a continuous line scale converts Hedonic measurements that otherwise cannot be analysed statistically into numerical data. Data were compared by analysis of variance (ANOVA) comparisons using R statistical software. 

Evaluation of Treatment Factors

Treatment factors for this experiment were total fermentation time, initial temperature of secondary fermentation and the type of fruit product used in secondary fermentation.

Initial fermentation time was conducted in 1L continuous batch reactors over a period of six days. The total fermentation time is then calculated by this initial fermentation time and time to complete the secondary fermentation. Literature states the optimum fermentation times of commercial products is around six to ten days [Bishop et al., sec.2]. However, the literature also states that fermenting kombucha for ten or more days results in a vinegar taste (caused by the accumulation of organic acids) that is off-putting to consumers. Any further fermentation time (>15 days) will result in dangerous levels of organic acids that are not safe for human consumption.

Secondary fermentation takes place after the initial fermentation where additional fruit juices are added to the fermented kombucha batch to add additional flavour profiles. This fermentation time period typically lasts between two to five days. As any longer fermentation time increases the bitterness attribute of the overall flavour profile. Once initial fermentation time was complete orange juice, concentrate (Syrup) and frozen orange pulp were added to the batches. Each fruit type had different initial temperatures ranging from frozen (-8℃) to chilled (4℃) to room temperature (20℃). These temperatures were chosen to either reduce or increase the temperature of the raw kombucha once mixed. Ideally generating the levels for the initial temperature of secondary fermentation. To ensure each sample can be statistically differentiated, samples were refrigerated at 3℃ after extraction. Therefore, limiting bacterial growth and thus fermentation before sensory trials.

Lastly, as previously mentioned there were three different types of Orange fruit products added to the raw kombucha during secondary fermentation. These types of products ideally have different adsorption ratios that will affect the amount present when analysing the final sweetness attribute.

Description of Experiment Planning

Participants:

Participants for the panel were general consumers (Male and Female) of kombucha products, who responded to advertisements. 

Requirements included individuals who weren’t pregnant or lactating, smoking or vaping, have an insufficiency of taste or allergic to caffeine or oranges. 

Training was provided for the sensory panel by Massey University’s Food Experience and Sensory Testing (FEAST) staff at the laboratory within a three day period, to ensure each panellist could adequately assess the range of kombucha samples. This taught assessors to accurately identify the sweetness sensory attribute of the kombucha and also to grade the attribute on a scale so that samples could be compared to each other. In return, participants were compensated with a supermarket gift voucher.

Ethics:

Due to the nature of sensory panels (research carried out on humans), a health and safety waiver was signed by participants and then approved by the Massey University Human Ethics Ohu Matatika 1 committee. 

Amount Of Participants Needed:

The sensory portion of this research was quantitated by the use of triangle and triad tests. Accordingly the amount of participants needed for this research was based on the minimum number of correctly guessed samples to reject the “Null Hypothesis”. Rejecting the Null Hypothesis refers to there being enough of a significance value to be able to state that the samples are statistically different from each other. Therefore, the number of participants were chosen by how many people can FEAST reasonably pay or reward and to meet the significance threshold. Therefore, if twenty-five individuals were to assess the samples a minimum of thirteen people would have to correctly guess the samples, to reach a significance value of 5%.

Randomisation:

Sensory:

Samples given to panelists were randomised in various ways to limit many different variables (errors) that indirectly affect data analysis. Firstly, randomised codes were used for each fermented sample. These codes were generated in R.studio using the rnorm command to limit bias. Thus limiting ranking errors of samples due to the scoring of samples based on imposed numerical scaling (i.e. 1st, 2nd or 3rd). Samples were also handed to panelists in random orders to limit order error. Therefore, each sample is represented appropriately for the attributes without contrast to the precursor sample.

Experiment:

As this experiment operates under set conditions for each batch, little can be randomised. However, the type of orange fruit product can be randomly allocated for each fermentation time. Therefore, limiting the random noise by generating a normal distribution of the treatment factors that ideally translate to the perceived sweetness attribute.

Replication:

Sensory: 

Replication was carried out by repeating triangle and tetrad tests with various sample variations before sampling fatigue occurred. Introduction of blanks in these tests were used to limit adaption. Adaptation error is where panelists become less sensitive to a stimulus over time, limiting a panelists ability to differentiate samples. Therefore, panelists were also encouraged to drink the water provided between testing samples.

Extraction of Samples:

Each extraction of raw samples was done in triplicate for each orange product. Resulting in three extractions of the raw kombucha at three different time periods and from these extractions a further three samples with each orange product. Therefore, generating data for all treatment factors individually.

Analysis of Proposed Data Collection

Blocking:

Blocking for this experiment follows a randomised complete block design (RCB). Where the total fermentation times (thirteen days) have been split into three orange products. Thus generating thirty-nine experimental units of perceived sweetness. An example of this blocking can be seen below in later sections, where all values are generated randomly.

Local Control and Covariance:

As not every source of variation from this experiment can be removed by randomisation, there must be variables or treatments that can be controlled locally. In this experiment the initial temperatures of the orange products is the only predetermined value. Other treatment factors are covariance as both the initial secondary fermentation temperatures and total fermentation times must be measured in terms of the initial temperatures of the orange products. Therefore, further analysis of covariance is needed when analysing the final data set.

Data Worksheet Set Up:

Below is a code block demonstrating how this hypothetical experiments data worksheet would be set up. Responses from these codes are random and the code itself will be used over all responses.

set.seed(2235)
Pulp <- rnorm(25, mean = 13, sd = 1)
Concentrate <- rnorm(25, mean = 13, sd = 1)
Juice <- rnorm(25, mean = 13, sd = 1)

Data_O <- data.frame(Pulp, Concentrate, Juice)
write.csv(Data_O, file="Orange_Product.csv", row.names=FALSE)
glimpse(Data_O)
## Rows: 25
## Columns: 3
## $ Pulp        <dbl> 11.00389, 13.76971, 13.79833, 12.80750, 15.33081, 11.95609…
## $ Concentrate <dbl> 12.65008, 12.91678, 12.74005, 15.12053, 12.25699, 14.06855…
## $ Juice       <dbl> 12.09084, 13.05600, 13.24473, 12.56414, 14.23931, 13.25725…
set.seed (1134)
Total_Fermentaion <- rnorm(25, mean = 10, sd = 1)
Inital_Temperature <- rnorm(25, mean = 10, sd = 1)

Data_F <- data.frame(Total_Fermentaion,Inital_Temperature)
write.csv(Data_F, file="Fermentaion.csv", row.names=FALSE)
glimpse(Data_F)
## Rows: 25
## Columns: 2
## $ Total_Fermentaion  <dbl> 10.219972, 8.504518, 11.046418, 9.069163, 10.742462…
## $ Inital_Temperature <dbl> 10.393375, 7.827722, 10.375653, 10.851830, 9.405089…

Graphs and Tables Used to Explore This Data:

Blocking:

set.seed(4444)

Time <- runif(25, min = 6, max = 25)

Fermentaion_Total <- data.frame(Time, Pulp, Concentrate, Juice)  |>
pivot_longer(2:4, values_to="Sweetness", names_to="Oranges") |>
glimpse()
## Rows: 75
## Columns: 3
## $ Time      <dbl> 24.714451, 24.714451, 24.714451, 8.062776, 8.062776, 8.06277…
## $ Oranges   <chr> "Pulp", "Concentrate", "Juice", "Pulp", "Concentrate", "Juic…
## $ Sweetness <dbl> 11.00389, 12.65008, 12.09084, 13.76971, 12.91678, 13.05600, …
Fermentaion_Total.lm<- lm(Pulp~Concentrate+Juice, data = Fermentaion_Total)
anova(Fermentaion_Total.lm)
## Analysis of Variance Table
## 
## Response: Pulp
##             Df  Sum Sq Mean Sq F value  Pr(>F)  
## Concentrate  1  3.9757  3.9757  3.3075 0.08261 .
## Juice        1  0.4255  0.4255  0.3540 0.55794  
## Residuals   22 26.4444  1.2020                  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Graphs:

Boxplot - An initial examination of the effect of orange products on the sweetness attribute. Used as a visual representation of the differences (not used in statistical analysis).

Data_Or <- data.frame(Pulp, Concentrate, Juice)
boxplot(Data_Or, main="Preceived Sweetness Ratings by Orange Product", ylab="Sweetness", col="lightgreen")

Interaction Plot - A statistical analysis of the effect of total fermentation time of all orange products, resulting in a range of sweetness attribute values. As the data for this plot was generated at random the graph is quite messy. However, as panelists will be trained on the detection of the sweetness attribute in kombucha, the real world graph should have similar sweetness attribute values between panelists.

library(ggplot2)
ggplot(data=Fermentaion_Total, aes(x = Time , y = Sweetness,color = Oranges, group = Oranges))+geom_point()+geom_line()+
  labs(title = "Interaction Plot For Sweetness",
       x = "Time (Days)",
       y = "Sweetness",
       color = "Oranges")

Below is an example of a pusedo real world interaction plot with only means of the orange product samples and total fermentation for comparison.

set.seed(5513)

Pulp_i <- rnorm(3, mean = 13, sd = 1)
Concentrate_i <- rnorm(3, mean = 13, sd = 1)
Juice_i <- rnorm(3, mean = 13, sd = 1)
Time_i <- runif(3, min = 6, max = 25)

Fermentaion_Total_i <- data.frame(Time_i, Pulp_i, Concentrate_i, Juice_i) |>
pivot_longer(2:4, values_to="Sweetness_i", names_to="Oranges_i")

ggplot(data=Fermentaion_Total_i, aes(x = Time_i , y = Sweetness_i,color = Oranges_i, group = Oranges_i))+geom_point()+geom_line()+
  labs(title = "Interaction Plot For Sweetness",
       x = "Total Fermentation Time (Days)",
       y = "Sweetness_2",
       color = "Oranges_i")

The tables used to explore this data are the same tables presented in the Worksheet Set Up section.

Initial Model For Response Variable

The inital modeling for the sensory measurements is the same as the previous section. Where the sweetness attribute was measured by the scaling in rating of samples and modeled by interaction plots and ANOVA analysis.

Validation of this model was completed by analysis of the second interaction plot. From this randomised data it was determined that the Orange Concentrate is not differentaible between the both samples at 19 and 19.5 days. However, the Pulp and Juice samples can be differentiated at all total fermentation times. Therefore, an intersection of two different lines (interaction) means that panelists could not tell the difference between samples (equally sweet) and thus score these samples around the same rating. Further validation is sorted from ANOVA analysis:

Fermentaion_Total_i.lm<- lm(Pulp_i~Concentrate_i*Juice_i, data = Fermentaion_Total_i)
anova(Fermentaion_Total_i.lm)
## Warning in anova.lm(Fermentaion_Total_i.lm): ANOVA F-tests on an essentially
## perfect fit are unreliable
## Analysis of Variance Table
## 
## Response: Pulp_i
##               Df  Sum Sq Mean Sq F value Pr(>F)
## Concentrate_i  1 0.46129 0.46129     NaN    NaN
## Juice_i        1 0.05469 0.05469     NaN    NaN
## Residuals      0 0.00000     NaN

Ideally, the ANOVA table should hint at differing significant values that relate to the interactions between samples. These values can be the result of various sensory, instrumental, experimental and/or sample errors or there is just no difference in sweetness.

Similar analysis and modeling was completed for the effect of total fermentation time and initial secondary fermentation time on the sweetness attribute. As the sweetness attribute can only be tested by human panelists, the outcome was the interaction of samples.

Description of Time Budget

Below is the initial activity calendar containing the target time frames that each step of the study will take. However, the calendar may vary dates due to real life events that are inevitable. Regardless of this variation, the extraction of kombucha samples will be within a week’s schedule. Allowing a relatively equal fermentation time between each sample, limiting errors. Alongside the calendar will be a sign off sheet for each task. Therefore, if there is any concerning variation in data, it can be linked back and discussed with a team leader.

Date Task
31/01/2025 Experiment is approved
03/02/2025 Advertisements for sensory testing begins
04/02/2025 Large Batch of Kombucha is made
7/02/2025 Secondary Fermentation and addition of Orange products
10/02/2025 First sets of samples are removed from the batch.
13/02/2025 Second sets of samples are removed from the batch.
20/02/2025 Third sets of samples are removed from the batch
19/02/2025 Panelists are chosen
21/02/2025 Panelists training
26/02/2025 Sample testing
28/02/2025 Analysis of sensory results
14/03/2025 Journal writing and peer reviews.

Other costs that effected this research was:

These costs may also effect the time budget in different real world ways.

Draft “Materials and Methods” Of Reference Articles

Materials:

Assessors:

Samples:

Equipment and other instruments:

Methods:

Samples were made by combining citric acid and either glucose or fructose and either volatile citral or limonene in mineral water. These samples were then carbonated with the use of two-way switch CO2 gas canister into the screw-topped glass vials. Lastly, the vials were refrigerated until needed for sensory testing.

Testing of the carbonated mineral water samples were tested under set laboratory conditions (room temperature, air conditioned room with natural sunlight). Samples (18, 38 ml) of 38 ml were presented to assessors in sets of three with 15 minutes breaks between each set. To ensure no carry-over effect, samples were presented within a minimum of 1 minute.

References and Citation Details

  1. Bishop, Peyton, et al. “Kombucha: Biochemical and Microbiological Impacts on the Chemical and Flavor Profile.” Food Chemistry Advances, vol. 1, Oct. 2022, p. 100025. https://www.sciencedirect.com/science/article/pii/S2772753X22000144#sec0004

  2. Batista, Patrícia, et al. “Kombucha: Perceptions and Future Prospects.” Foods, vol. 11, no. 13, July 2022, p. 1977. PubMed Central, https://doi.org/10.3390/foods11131977.

  3. Pearce, Sarah. Kombucha Second Fermentation: The Home-Brewer’s Guide. 11 Mar. 2023, https://buchabuddha.com/kombucha-second-fermentation/.

  4. Villarreal‐Soto, Silvia Alejandra, et al. “Understanding Kombucha Tea Fermentation: A Review.” Journal of Food Science, vol. 83, no. 3, Mar. 2018, pp. 580–88. DOI.org (Crossref), https://doi.org/10.1111/1750-3841.14068.

  5. Wang, Boying, et al. “Kombucha: Production and Microbiological Research.” Foods, vol. 11, no. 21, Oct. 2022, p. 3456. PubMed Central, https://doi.org/10.3390/foods11213456.

  6. Hewson, L., et al. “Gustatory, Olfactory and Trigeminal Interactions in a Model Carbonated Beverage.” Chemosensory Perception, vol. 2, no. 2, June 2009, pp. 94–107. Springer Link, https://doi.org/10.1007/s12078-009-9043-7.

AI Statement

I, Rose Matiu, hereby confirm that I used no generative AI tools or systems in the completion of this assessment. I confirm that the document I have submitted was entirely written by me.

I acknowledge that any undeclared use of generative AI will constitute academic dishonesty and will be dealt with according to relevant University policy.

I understand that I will be held accountable for any academic misconduct that arises in breach of any relevant University policy, as well as the consequences of such infringements.