external image airbnb-logo-293-5b1924f36d180a53fdca602da3e5bc6c.png

Done by Kai Wei Tan

Introduction


Before you read the below wiki, let us try a simple exercise. Imagine you want to go on a holiday trip to Singapore. Below are some housing options for you to choose. There are 4 options. The left hand corner with a price of $40, simple furniture layout and only 2 reviews. The right hand corner with a price of $44, slightly sophisticated furniture layout. and more than 20 positive reviews The bottom left hand corner with a fairly high price of $113 and well furbished layout but no review. Lastly, a newly furnished room with a price of $44 and 5 reviews. Which one will you choose?

Screen Shot 2015-11-27 at 3.38.15 PM.png
Airbnb


A simple exercise conducted on a few of my friends turns out that 90% of them will choose the top left hand corner and the bottom right hand corner. After the simple survey, I asked for their feedbacks on what factors affected their decision. And the answers are not surprising: the review and the price. I continue to probe them and ask: Does the tone of the picture affect your decision?

Majority replied that the tone of the picture affected their decision. After asking them to review their decision making process, most of them decide to choose the top right hand corner, taking into consideration of the number of good reviews. So, is it true that the tone of the picture unconsciously bypasses other factors such as review and price in their decision making process?

Background



So how does the above simple survey relate to my project?

To be successful in any business, we have to know what the consumers are thinking. If we can understand how our consumers think, we can curate better products for them. In general, consumers choose to rent based on the price, location, pictures and the reviews. What if we can identify the weightage of each factor in deciding how much it influences the decision of the consumers. With the wealth of data collected over the years, Airbnb can exploit the Big Data to shed more insights into the behavior of the consumers and ultimately find out not just the factors that influence their behavior but also quantify the factors in their decision-making process. For the purpose of this exercise, we will look specifically at the power of graphic visualization in determining user's decision.

test.jpg
Natural Lighting vs Dark Lighting



PHAM


Problem: Some landlords(refer to users who list their house for rent) find that their housing options are not picked by users even though they offer one of the best deals in town, matching in terms of price, location and review. However, given the increasing competitiveness of Airbnb(with more and more users joining the community and renting out their place), how can landlords further differentiate themselves from the crowd, at least in the short to medium term?

Solution: To bring in more revenue for Airbnb, we have identified a simple strategy for Airbnb; advise users to post pictures of house under natural lighting.

Hypothesis: Brighter and natural lighting attract user's attention, or at least, drives their first click. Interestingly, psychology and academia have found that we have a higher preference for bright color rather than dull color. As aptly put by R Douglas Field, in his article, Why We Prefer Certain Colours, "Color preferences are deeply rooted emotional responses that seem to lack any rational basis, yet the powerful influence of color rules our choices in everything".

Action: To conduct a rigorous approach, rather than a simple survey which I have conducted above, we will look no further than random assignment. The definition of random assignment is as such:

"The key to randomized experimental research design is in the random assignment of study subjects -- for example, individual voters, precincts, media markets or some other group -- into treatment or control groups. Randomization has a very specific meaning in this context. It does not refer to haphazard or casual choosing of some and not others. Randomization in this context means that care is taken to ensure that no pattern exists between the assignment of subjects into groups and any characteristics of those subjects. Every subject is as likely as any other to be assigned to the treatment (or control) group. Randomization is generally achieved by employing a computer program containing a random number generator. Randomization procedures differ based upon the research design of the experiment. Individuals or groups may be randomly assigned to treatment or control groups. Some research designs stratify subjects by geographic, demographic or other factors prior to random assignment in order to maximize the statistical power of the estimated effect of the treatment (e.g., GOTV intervention). Information about the randomization procedure is included in each experiment summary on the site."
Source: Yale Institution for Social and Policy Studies

After random assignment of users to a page, we can conduct A/B testing. We can randomly assign housing of similar characteristics except for the display picture with varying color tone side by side. We will then look at the actual measurement to see the validity of our hypothesis.

Actual Measurement: There are 2 main aspects of the measurement: the first click and the conversion rate. We want to know if brighter picture will result in a first click. Next, the users will most likely compare their current choice with the next option. As such, we want to know if the brighter picture does lead to a conversion of booking. In short, we will look at the (1)percentage of first click, (2) within those first click, how high is the conversion rate (3) even without first click, how many percentage of users change their options to the brighter picture.

Evaluation: We have to be careful in our impact assessment. Typically, the increase in the percentage of first click is what we identified as the "compliers" and "always takers", while the rest who did not respond as "defiers" and "never takers". Compliers are those when "treated" will respond to the outcome of the hypothesis. In this case, those people when treated with a choice between natural lighting picture and darker picture, they will choose the natural lighting picture. Always takers are those regardless if they are given a choice, they will always choose the natural lighting picture. Defiers are those that will choose natural lighting if not given the treatment, however, upon given the treatment(the choice to choose), they will not choose the natural lighting. Lastly, never takers will never choose the natural lighting picture regardless whether they are given a choice. Usually in statistical studies, "compliers" take up around 10 percent of the profile. However, as Airbnb seeks to attract more people we have to weigh in to compare whether "defiers" are more or less than "compliers". Intuitively, we would not expect "defiers" to be more than "compliers". Nevertheless, we want to capture as many "compliers" as possible and to go even further to change the minds of the "never takers". Given the nature of the hypothesis, we will not go that far. In short, this project seeks to capture our main target audience, the "compliers".

TABLE1.png


Limitation


However, due to the setup of the page, Airbnb can not just freely setup the page according the PHAM setup mentioned above. To be more precise, it will indeed be an administrative nightmare to ask users to upload both pictures of the same setting but under different lightings. Instead, we can conduct preliminary analysis. With the advancement in Machine Learning, in particular Computer Vision, (Computer vision is a field that includes methods for acquiring, processing, analyzing, and understanding images and, in general, high-dimensional data from the real world in order to produce numerical or symbolic information, e.g., in the forms of decisions) we can sort out sub sample of data when users have experienced two pictures of varying tones but with similar characteristics(i.e reviews, price and location).

Granted, we do not have a perfect random assignment but we have "as good as a random assignment". We can then look at the click rate to evaluate our hypothesis.


external image img2.png


Conclusion


If indeed the hypothesis is true, Airbnb can just recommend end users, in particular the landlords to upload pictures of their house under natural lighting. Having sought comments from peers, I received most comments that they do see the importance of visual image, however, they seemed to be skeptical about the magnitude of natural lighting on their purchasing behavior. Granted, natural lighting may be one of many factors, and could be very well be insignificant, in their decision-making. However, this project main highlight is to dig into the treasure trove of data and find seemingly trivial factor yet profound factor in guiding consumer's behavior. Most essentially, this simple exercise can be extrapolated to many other exercises to shed more behavioral insights of users.


external image behavioral+psych+-+beh+economics.jpg




References


Consumer Reports

http://www.consumerreports.org/cro/magazine-archive/2011/june/shopping/product-packaging/overview/index.htm

Yale

http://isps.yale.edu/node/16697#.VltPZ9-rRhE

Wikipedia

https://en.wikipedia.org/wiki/Computer_vision

Computer Vision

http://www.blog.namar0x0309.com/wp-content/uploads/2014/07/img2.png

Local Average Treatment Effect

http://egap.org/methods-guides/10-things-you-need-know-about-local-average-treatment-effect

Image References:

AIrbnb

https://www.airbnb.com/s/Singapore?guests=&checkin=11%2F28%2F2015&checkout=11%2F30%2F2015&ss_id=3fedy5cj&ss_preload=false&source=bb

Flickr

https://www.flickr.com/photos/28067431@N03/6912021784

Behavioral Insight

http://static1.squarespace.com/static/5315307fe4b04a00bc137c49/531dc3dbe4b0ddc80cd057c5/5331e95be4b088cc57b08840/1395780065146/behavioral+psych+-+beh+economics.jpg