Results of this randomized controlled trial showed that a 25% discount on fruits and vegetables was significantly associated with higher total fruit and vegetable purchases in a web-based supermarket. The results showed that, after appropriate adjustments, the experimental group purchased 984 g more fruits and vegetables for their household for a week than the control group, which indicates a 25% difference. This difference points to a price elasticity (PED) of 1.0 and was independent on scores on habit and price perception. Also it was revealed that the discount on fruits and vegetables neither lead to higher expenditures in other (unhealthier) food categories nor to a higher total amount of calories purchased. These findings could have important implications for public health.
One rationale for introducing food pricing strategies is that monetary costs of a healthy diet may form an important barrier for low-income consumers in adopting such a diet . Numerous studies have shown that nutrient-rich, low-energy-dense foods (e.g., fruits and vegetables) are generally relatively more expensive than high-energy-dense, fat and sugar rich foods [32–34]. In addition, it is suggested that in the current market, fruit and vegetables are promoted less than more profitable, highly processed foods containing more fats and sugars [35, 36]. Since different studies have shown that, especially for low-income consumers, price is a major factor in food choice [37–39], pricing strategies are promising in stimulating healthier food alternatives. Already, marketing research has indicated price as a key tool in directing consumer behavior .
So far, the evidence on the effects of food pricing interventions was mostly restricted to interventions in smaller environments such as vending machines or work-site cafeterias. To our knowledge, our study is one of the first experimental studies on the effects of discounting fruits and vegetables in a virtual supermarket environment. When our results are judged against comparable studies, our findings are similar. First, The New Zealand SHOP study found that a 12.5% price reduction of healthier foods lead to 11% more healthy food purchases . Also an economic modeling study by Jensen and Smed found that reducing VAT on fruits and vegetables from 25% to 12.5% lead to an increase in sales of 8% of those products . Finally, French et al. conducted an experiment in high-school canteens and found that a 50% discount on fruits and baby carrots lead to a fourfold and twofold increase in sales respectively . All together, there is increasing evidence that lowering the prices of fruits and vegetables is effective in stimulating the purchase of these foods. Recently, Andreyeva and colleagues published a review on the PED of food. Based on a selection of 160 studies, they concluded that food is elastic and that the highest PED was found for food away from home (restaurant meals and fast food), soft drinks, juice, meats, and fruit .
Nevertheless, there are also studies reporting possible negative side effects of subsiding healthier foods. For example, a study by Epstein and colleagues on a purchasing task in a laboratory setting found that discounting healthy foods with 12.5% or 25% lead to an increased number of total purchased calories since respondents did not only increase healthy, but also unhealthy food purchases . A following relevant consideration regarding the effects of lowering fruit and vegetable prices is that people may purchase more of those products additional to their regular purchases instead of replacing other products by fruits and vegetables. In our study, we did not find that people spent the money they saved from the discounts in other (unhealthier) food categories. Also we found that both groups purchased similar amounts of calories and a similar number of products. An explanation for this difference in findings may be the studied product assortment. In Epsteins' study, people were able to choose between 30 healthier and 30 unhealthier products, whereas our web-based supermarket had a variety of 512 products. In addition, we only discounted fruits and vegetables whereas Epsteins' study discounted a wider range of healthier products . This means that a fruit and vegetable subsidy may have better overall effects on food purchases than a discount on all healthier foods. Nevertheless, it is important to study this compensation effect carefully in experiments in real supermarkets, under different circumstances and by incorporating overall household expenditures (also outside the supermarket).
Another important aspect is that our results may be an underestimation because the discounts in the web-based supermarket were silent. Normally, when products are discounted, effort is made to draw people's attention by using signs or advertisements. Previous authors have suggested that people have a poor reflection of prices  and by using additional strategies; people become more aware of the discounts. Also, people have the tendency to buy a product simply because it is on sale [43, 44].
The results of our study indicate that a discount on fruit and vegetables is effective in stimulating purchases of those products. Still, our study found only significant effects on fruit and vegetables combined and not for fruit or vegetable purchases separately. Nevertheless, the separate effects (+504 g vegetables and +481 g fruit per household per week) were also quite large and are considered relevant. These numbers point to a difference of 29 g and 28 g per person per day respectively. The latest Dutch Food Consumption Survey (2007-2010) showed that adults in the age 30 - 51 consumed a daily average of 121 g of vegetables and 77 g of fruit . Increasing these numbers up to recommended levels of 200 g of fruit and vegetables per day could have large implications for public health . An explanation for the non significant results, however, can be found in a lack of power. The used standard deviations in the power calculation were much smaller than the standard deviations found in our study. Therefore, a larger sample than expected was required to find significant results. It is therefore important to study the effects of fruit and vegetable price discounts in a larger sample. Such a study is also vital to gain more insight into the effects for specific groups, such as people with a low income or for ethnic minorities. Financial barriers against buying sufficient fruits and vegetables principally apply to low-income groups [10, 11]. In our study, a majority of study participants had a standard income or above, making that their income was relatively high. Nevertheless, our results indicate that discounting fruits and vegetables was effective in this relatively high income sample as well, meaning that it can be expected that this strategy is equally (or even more) effective among people having limited financial recourses. Finally, results can not be directly generalised to populations with different eating habits and a different culture as opposed to the Netherlands (such as other EU countries or the US). Nevertheless, seen the generally low fruit and vegetable consumption in the entire EU  and also in the US  it can be expected that lower fruit and vegetable prices can have similar (or even greater) effects there as well.
A strong merit of our study is the use of the three-dimensional web-based supermarket which closely images a real shopping experience. Nevertheless, the assortment of the web-based supermarket is not as extensive as a real supermarket. Also, the Virtual Supermarket does not give insight into how people may shift to non-food items as a consequence of the price changes. Besides, the results are limited to a supermarket environment and do not give insight into effects at other point of purchase settings. Nevertheless, people buy most of their food at supermarkets (Dutch supermarkets' market share in 2011 was 86% ) and this seems thus the most obvious environment for interventions. Another limitation is that people may react differently in a real shopping situation with real products and real money compared to our web-based situation. Still, a large majority of the participants stated that their purchases in the web-based supermarket resembled their regular food purchases. Also, participants who had trouble in understanding the application were excluded from analysis. Furthermore, there is evidence that peoples' virtual behavior largely corresponds with their actual behavior. Sharpe et al. (2008) validated meal and beverage choices made in a virtual road trip survey by comparing those choices with choices made in a real McDonalds a week later. The authors found that peoples' simulated purchase behavior is highly predictive of their actual behavior . Moreover, compared to previous studies where a supermarket environment was modeled using only 60 products  or using online drop-down lists , our three-dimensional, 512 products containing application seems a good quality research instrument. Unlike this, it is important to validate our results in a real shopping environment. A final limitation of our study is that some selection bias may have occurred because participants were self-selected. Still, participants were not aware of the research aims and were blinded with regard to assignment of the research conditions, which is considered a merit of our study.