Modern Metrix

Color Forecast

This is awesome! French clothing company Pimkie has installed cameras on the streets of Milan, Paris and Antwerp (supposedly, Europe's most fashion-conscious cities) to capture images of people's clothing. The captured data is then converted into real-time "color forecast" on their webisite, so you can see what colors are trending right now in these cities. You can also look at the colors that were trending yesterday, last week, and last month. For example, these were the most popular colors in Paris in past 30 days:

Imagine taking this concept further, by capturing gender and age of passerby's to further enhance the forecast? Or capturing the images of shopping bags on the streets  and then projecting which stores are currently trending, on the giant screens, or on websites and in mobile apps?

Gmail Meter Helps You Analyze Your Gmail Usage

Just as I started looking for better ways to measure my Gmail usage, Gmail labs rolled out the Gmail Meter script! Among the statistics, provided on a monthly basis and summarized via nice charts, are volume, traffic patterns and time analysis, length of messages and conversations, and even how long does it take for you to respond to e-mails. Check it out!

Quantifying "Walkability"

Slate looks at how small start up Walk Score is attempting to quantify the "walkability" of geographical locations. 

Where Facebook Ads Work Better: States with the Highest and Lowest Click-Through Rates

This article in Mashable shares interesting statistics on geographical difference in click-through rates for Facebook ads,  provided by digital marketing agency:

"Facebook users in South Dakota, Tennessee and Colorado click through ads on Facebook at record rates. In South Dakota, Facebook users click on ads at a rate of 174% the national average.

Meanwhile, residents of Vermont, Alabama and Indiana are the least likely to take the bait. In Vermont, the click-through rate (CTR) is 57% the national average."

The differences are not explained. If you ask me, the first what comes to mind is demographics, most likely, education (states with less educated population have higher click-through rates). I also suspect that states with the higher click-through rates are are ealier stages of Facebook adoption (the novelty factor comes in play, hence more clicks). But that is just IMHO.

Free Twitter Data!

I am always on the look out for free data to play with, and today stumbled upon awesome depository of Twitter data on Infochimps website. Check it out!

 

Neighborhood Lifestyle Metrics using Foursquare Data!

In this post on their blog, Foursquare engineering team describes how they used Foursquare activity data to compute lifestyle metrics of city neighborhods, and then used these metrics to compare neighborhoods across different cities:

"First, to collect data about neighborhoods, we built some Hive queries to access our large collection of check-ins (stored in S3) and count the number of check-ins per category for every neighborhood in the world. For example, the East Village of New York has 230k check-ins at bars, 57k check-ins at pizza places, 18k check-ins at yoga studios, and 34k check-ins at karaoke places (hipsters like to sing!).

We then used MATLAB to represent each neighborhood by a 400 dimensional vector which specifies the normalized probability distribution of checking in to a place in each category relative to the baseline distribution of the city. This approach allows us to compare neighborhoods with each other using a similarity metric such as cosine similarity, or KL divergence."

What an awesome hackday project!

Impact of Survey Length on Measurement Quality and Participation Rates

This interesting study by Lightspeed Research looked at the relationship between the online survey length and quality of measurement and participation. In summary, study found that longer surveys result in higher drop out rates, and less reliable measures of constructs of interest. Interestingly, there was relationship between certain demographic and probability of dropping out, although that did not necessarily affect the representativeness of the overall sample:

"Respondents who drop out are somewhat different than those who complete the survey. Demographically, they are likely to be older, retired, male and from smaller households without children. In addition, those who drop out are less likely to do all the grocery shopping and tend to purchase snack foods less often. Despite their differences, the dropouts are so few in number that their exclusion does not harm the representativeness of the sample."

The study also found that, possibly, due to the fatigue, the quality of certain measure was affected by the survey length: 

"In the Lightspeed Research study, the concept scores for the 8-minute questionnaire tend to be higher than scores for the longer questionnaires. The 20- and 24-minute questionnaires receive fewer “definitely would buy” ratings compared to the 8-minute version. There are also several notable differences on liking, value, uniqueness, and believability between the 8-minute version and the longer versions (see Table 2). Also, several important differences occur between the 8-minute questionnaire and the two versions with the extra upfront questions."

The position of questions pertaining to the main concept in the survey relatively to other attitudinal and demographic questions may also result in biased and poor quality responses. 

Supermarkets Introduce Scoring Systems for Food Products

It looks like more supermarkets and grocery stores chains start using the nutritional content scoring systems for food products, aparently, following the lead of Whole Foods (WH has been using their own ANDI scores for awhile now):

"A growing number of grocers are signing up to have the foods they sell evaluated and ranked for nutritional content at a time when government officials are studying whether more regulations are needed on packaging.

Two of the largest companies, NuVal and Guiding Stars, license their scoring systems to more than 3,000 grocery stores nationwide, according to officials at the companies.

The scoring systems have some differences. Guiding Stars uses a system of three stars to signify which foods are healthiest. Three stars is the best ranking. NuVal uses a 100-point system: The higher the score, the better the nutritional content."

While I agree that a single-scale scoring approach has obivous advantage compared to looking at calories/ carbs/protein/fat content separately, I am somewhat cautios about how these scales are normalized and validated. If the scale is normalized using a regular 2,000-calories daily requirements, and I am a fan of a low-fat low-carb high-protein diet (that is, my protein intake is higher and fat/carbs intake is lower than recommended by 2,000-calories diet), then this metric won't work well for me. The same would be true for follower of other kinds of specialized diets (e.g., Atkins).

Twitter to Start Selling Archived Tweets

One of the problems I encountered while trying to measure media users' engagement with content on Twitter was that data downloaded via Twitter API comprised only tweets dating back to only 30 days. As a result, any metrics that was based on behavior had to be defined as "in last 30 days" . Well, at last some great news: Twitter partnered with British start-up Datasift to sell access to tweets posted in last 24 months. 

Now that we have more historical data available, will someone finally start working on a "Census" of Twitter users, to give us probability-sample based estimates of basic behavioral metrics of entire Twittosphere, like average number of tweets and followers, click-through rates, etc.? 

New York City Council Frees NYC Public Data

This is great: NY City Council passed yesterday the Intro29A measure that requires all city agencies to make their public data available on NYCOpenData portal:

"Intro 29 would phase in over the next six years, starting with a technical manual to be released by DoITT this summer guiding city agencies on how to format and upload their data. Agencies would have until early next year to upload all public datasets currently on the internet to the new portal. By 2018, all public data would have to be posted. For the time being, DoITT is planning to power the new data portal with Socrata, the service that currently hosts NYC OpenData."