Data Cleaning and Analysis

The first step of data cleaning was to remove respondents who had never lived in Sweden and thus were not familiar with Swedish meat prices and customs. For this, the answer "Do you currently live in Sweden" served as a gatekeeper question. Only the answers of those living or having lived in Sweden were considered. Furthermore, it was confirmed that those indicating to have never lived in Sweden also indicated a different nationality, as to ensure that it was not a simple filling-in mistake and that their responses were in dollar or euro values as opposed to SEK. Hence the entries 14, 26, 49 and 62 were removed.

Next, the second gatekeeper question "Do you consume meat" was used to filter out vegetarian/vegan respondents. This was since they might have a skewed perception of meat prices, or what their value should be. A Van Westendorp analysis usually asks the purchase likelihood and consumption of a good too, which this question served for.

Also, the thesis aimed to analyse meat consumers response to prices, making them not part of the target population. This resulted in the elimination of additional 22 respondents, leaving 81 valid responses. One set of duplicate answers (Observation 51

due to lack of or faulty information. Three of these respondents left a comment that they were not aware of meat prices. In the end, 75 responses could be used for the analysis.

Of the remaining respondents, 79% were currently living in Sweden, and 21% had in the past. 53% had the Swedish nationality, 14% were Austrian, and the remaining 33%

were from countries such as Germany, France and Finland.

The income range of the respondents’ households ranged from less than 10.000SEK/month to more than 100.000SEK/month. The median income per household was 30.000-40.000. On average, the households consisted of two individuals. Per household two income-generating members were the norm. The income average per capita was around 22.600SEK, which was slightly below the average Swedish income of 2019 (Statista, 2020). The age range of participants was from 20-29 to 70-79, with most participants, however, being in the 20-29 age group, as can be seen in Figure 18.

The gender distribution was 58% female to 40% male, and one person identified as

"Prefer not to say".

Figure 18: Age distribution of questionnaire participants (cleaned data).

Some participants noted that their data had been submitted in kilogram price instead of 200g prices. Hence the data had to be modified before being useable. In order to ensure that outliers do not compromise the data, the data group had to pass an outlier test. First, however, to determine the correct outlier test, normality tests were performed. All individually missing data were replaced by the mean, not to lose interpretable data in

Age distribution

20-29 30-39 40-49 50-59 60-69 70-79 Blank

the process. All four tests rejected the Null-Hypothesis (H0 – The data is normally distributed), indicating that the data was not normally distributed.

Variable\T

a bargain? <0,0001 <0,0001 <0,0001 <0,0001

get exp.? <0,0001 <0,0001 0,004 <0,0001

too exp.? <0,0001 <0,0001 <0,0001 <0,0001

too low? <0,0001 <0,0001 <0,0001 <0,0001

Pork

a bargain? <0,0001 <0,0001 <0,0001 <0,0001

get exp.? <0,0001 <0,0001 <0,0001 <0,0001

too exp.? <0,0001 <0,0001 <0,0001 <0,0001

too low? <0,0001 0,000 <0,0001 <0,0001

Poultry

a bargain? 0,000 <0,0001 0,000 0,006

get exp.? <0,0001 <0,0001 <0,0001 <0,0001

too exp.? <0,0001 <0,0001 <0,0001 <0,0001

too low? <0,0001 <0,0001 <0,0001 <0,0001

Table 5: Normality testing for the pricing questions leading to the conclusion that the data is generally not normally distributed.

However, the data appears to be sufficiently normally distributed when visually represented in histograms. Using the Grubbs Test for outliers indicated that the results are still useful. Furthermore, the later employed Van Westendorp method does not depend on the normality of data.

Figure 19: Histogram analysis example for observing the normality of the data.

The Grubbs Test identified outliers, which were then modified if it was clear that data was given in the wrong unit or removed if the data was deemed not useable. The price range for the first question was from 3.5 to 100, had a mean of 38.441 and a standard deviation of 21.127. The p-value of the Grubbs test amounted to 0.192. Hence H0 (there is no outlier in the data) cannot be rejected. The outliers are identified by the Z scores and given in Figure 20 below.

Figure 20: Outlier identification for the question of beef prices being a bargain, computed by using z-scores.

There might be two explanations for the outliers and price discrepancy: 1) a difference in meat quality bought, and hence a different accepted price level or 2) people misperceiving the questionnaire and answering in kilogram instead of 200g prices.

However, as the second is more likely, especially considering that some participants pointed this out to the author, it was surmised that the data might be modified into 200g

0

...a bargain - a great buy for the money?

Histogram (...a bargain - a great buy for the money?)

-4

prices. This is an excellent example of the limitation of a quantitative survey, which could have been avoided in a qualitative setting or direct interaction with the respondents.

Next, the data for the Van Westendorp method had to be validated. For this, all answers had to be analysed as to whether they had been filled in correctly, as displayed in Figure 21 (Pritchard, 2020). Answers that failed this test were removed uniquely for this type of meat, as to lose as little data as possible.

Figure 21: Data validation for the Van Westendorp method, as suggested by Pritchard (2020).

Next, the Van Westendorp method could finally be applied. The outcome is a graph indicating an acceptable price range for each type of meat; the optimal price point amongst the respondents, and the indifference point. The acceptable price range for beef for instance ranges from 20 to 65SEK for 200g, and the optimal price point is 36.7SEK and the point of indifference 40SEK. This is visualised further in Figure 22.

Figure 22: Van Westendorp price sensitivity meter output for beef prices in SEK.

A summary of the different price ranges and price points are presented in the following table:

Lastly, some questions regarding the respondents awareness of environmental and health issues were asked, as displayed in Figure 23, and whether they were willing to

IDP=40

Cheap / Expensive Not cheap / Not expensive Too cheap / Too expensive

Price Range Optimal Price Point

Table 6: Van Westendorp analysis summary: Price Range, OOP and PI for part 1 and 2.

change the price as a result of that knowledge: 53% are not willing to pay higher prices to compensate, whereas 47% are.

Figure 23: Awareness amongst respondents of the damage meat causes to health and the environment.

The Pearson correlation test was applied to see if any pattern could be found between the price points given between those participants who were more and less aware of the environmental and health concerns. The following scatter plot was produced for beef consumption in relation to awareness:

0 5 10 15 20 25

1 - Totally

Unaware 2 3 4 5 6 7 - Very Aware

Awareness of environmental and health impacts of meat amongst respondents (abs)

Absolute frequency envr Absolute frequency health

It is interesting to note that those with higher awareness, especially of the environmental damage, were also somewhat willing to pay higher prices before being reminded of the impact meat consumption has. The same is however less apparent for chicken and pork, the latter even having some negative correlation. For their scatterplots, please refer to Appendix B. Figure 24: Scatter plot for correlations between beef prices and meat externality cost awareness.

Also noteworthy: out of 35 respondents, replying to the question "Would you consume more meat at a fairer (higher) price?", 60% indicated that they would consume the same amount; 37% indicated that they would consume less, and one individual did state that they would consume more. The discussion of these results will be elaborated on in Chapter 5.