Can Algorithms Break the Law?

Here’s a breakdown of how 2 of the most prominent search algorithms on the web appear to be presenting information that it would be illegal for you or I to communicate.

  1. The first is the Twitter search algorithm, tweaked recently in a way which (inadvertently) increases the likelihood of people finding out ‘illegal’ information.
  2. The second is a Google algorithm, and 2 user interface changes which (again inadvertently) very much increase the chance of ‘illegal’ information being communicated.

imageobscured犀利士
4″ />

The Background – a Very Sad Story

In the entire history of the British legal system, there are apparently only four prisoners who have been given new identities. One of them was a boy (now a man) who used to be called Jon Venables. When he was a child, Jon Venables & another boy (Robert Thompson) killed a very, very young boy (James Bulger). If you live in the UK, you will almost certainly know this.

Because of their crime, Jon Venables & Robert Thompson are very likely to be in extreme danger if the general public can identify them. Therefore, they were given new identities, and those identities were protected.

Since Jon Venables was given a new identity, nobody is allowed to know his new name: It is illegal to publish anything claiming to identify him, or even to ‘purport to identify’ anyone as him whether it is him or not.

jvindie

Photos of Jon Venables were apparently published on Twitter recently. This hit the front pages of most newspapers in the UK (including, above, The Guardian).

The Twitter Search Algorithm Change

Until recently, Twitter’s search algorithm only returned very recent tweets. A few weeks ago, that was altered to also include much, much older tweets. That coincided with the 20 year anniversary of Jon Venables’ crime. As a result, lots of very old tweets and photos claiming to ‘out’ Jon Venables identity suddenly became visible far more easily, at a time when many were searching for his name. Many of these photos were retweeted; some new ones were posted, some were taken from Twitter and posted on other websites, and many users posted names that they claimed were Jon Venables’ new identity.

That hit the front page of most UK newspapers (though none made the link with Twitter’s algorithm update), and – of course – an investigation was carried out.

Twitter Search Last Week: Jon Venables

Here’s what happened if you searched Twitter for ‘Jon Venables’ a week ago:

venables

You can see there, Twitter has pulled a ‘Top news’ box covering the investigation. It has automatically pulled a ‘Top’ tweet that was Retweeted 49 times, claiming to out Jon Venables’ new identity & urging readers to spread it.

There are 2 big red blocks there too where I’ve obscured 2 important items:

Item 1: Top Photos

On the left (item 1) I’ve obscured 2 photos showing an adult man, roughly the age Jon Venables would be now. Those were posted by users, but the algorithm picked them out as ‘top photos’.

Item 2: A man’s name

Across the top (item 2) I’ve obscured a man’s name, flagged as being a ‘Related search’ for ‘Jon Venables’. Again, as with the photos, this was posted by Twitter users, but Twitter’s algorithm picked it out & highlighted it as being a related search.

Clean Up?

Since this all came out, the UK Attorney General has announced that there will be ‘contempt’ proceedings launched against people who posted images ‘purporting’ to be Jon Venables.

Twitter has obviously made a substantial effort to clean up the photos too.

Twitter Search Now: Jon Venables

Today, if you search for ‘Jon Venables’ on Twitter, the photos on the left (which previously showed an adult man) have gone. BUT, the ‘Related’ search term, showing a man’s name, is still in place:

jv2

If you click through on the man’s name in the obscured ‘Related Search’ up there, it leads you through to this:

venablesnew

Twitter Summary

In the first instance, Twitter’s algorithm automatically highlighted photos & details of an adult man whenever users searched for ‘Jon Venables’. Twitter users themselves posted the content, but the algorithm crowdsourced from that to highlight particular photos, and a particular name.

Twitter have obviously gone to some lengths to clean things up here. BUT, on searching for ‘jon venables’, the algorithm still leaves a fairly prominent trail toward a man’s name, and toward tweets linking to photos of an adult man.

And Google?

The Google story is much, much shorter. It involves 2 relatively recent additions to Google’s user interface:

  1. The ‘knowledge graph box’ – an area on the right of search results that is intended to reveal ‘Facts’ related to searches.
  2. Google’s updated Image Search results, which used to show images in the context of a web page, but now simply show the image on a black background on Google itself.

Here’s what happens when you search for ‘Jon Venables’ on Google today:

jvgoog

I’ve obscured an area on the right there (within the ‘knowledge graph’ box), where there are 2 photos (within a single image) of an adult man.

And when you click on that obscured area, it leads us to this quite scary screen:

jvred

A few weeks ago, clicking the image from search results would have taken you to the website in question (faded in the background), showing clearly that it was a website other than Google publishing it. Today, doing so keeps you on Google’s own property (note the ‘google.co.uk’ URL). It simply shows a black background, 2 photos of a man, and the name ‘Jon Venables’, the old name of a person it is illegal to identify.

10 Google Analytics Site Speed Tips

Google Analytics has a set of ‘speed’ reports, found under the the ‘Content’ left-hand navigation item. Here are 10 very quick “Google Analytics Site Speed” tips, collected together for ease of reading:

  1. All of the load speed data in Google Analytics is taken from users’ browsers themselves, so is fairly accurate.
  2. The browser stores all this speed data; Google Analytics then pulls that data at the point the code fires.
  3. Some browsers don’t support page load speed recording (notably Safari) so be very careful on ‘mobile’ speed.
  4. It’s tempting to pay attention to the top level “Avg. Page Load Time (sec)” metric that Google Analytics gives you. Slightly more useful than that: The ‘Distribution‘ report is tucked away behind a tab in the ‘Page Timings’ report. The distribution report is very useful, as you can see ‘% of pages loaded in less than X seconds’.
  5. ‘DOM Timings’ reports were added long after the main speed reports. They’re found in a similar area to the ‘distribution’ report, and are useful as they show how long it takes for your pages to become usable. (giving you ‘Avg. Document Interactive Time (sec)’ and ‘Avg. Document Content Loaded Time (sec)’ for each page).
  6. A big caveat: Google Analytics only tracks ‘page load speed’ for between 1-5% of pageviews as standard. You can fix that (see point 8).
  7. Because the sample size is so low, remember that ‘average’ page load speed is therefore affected if the speed of a page has only been recorded a few times.
  8. To improve the sample rate issue, you can add some extra code to your Google Analytics implementation. This essentially tells Google Analytics ‘try and collect page load speed data for 100% of page views’. Based on the earlier caveat that not all browsers support it, and the caveat that Google cap the total number of sessions they will record this for, you will never see a 100% sample rate, but if you are below that cap, this will increase it significantly. I have known sites achieve site speed data for 30% or more of pageviews, vs the default 1%. The code required to do this is as follows:
    • For older versions of Google Analytics (these will appear in your code as ‘ga.js’ or ‘dc.js’): _gaq.push([‘_setSiteSpeedSampleRate’, 100]); // this should go above the ‘trackpageview’ line in your Google Analytics code to improve accuracy
    • For newer versions of Google Analytics (these will appear in your code as ‘analytics.js’): ga(‘create’, ‘UA-XXXX-Y’, {‘siteSpeedSampleRate’: 100});
  9. If you’re looking for a good KPI on site speed, ‘% of pages loaded in less than 3 secs’ is a safe bet. (and ‘less than 7 seconds’ as a secondary).
  10. You may want to isolate that KPI further: ‘UK Only: % pages loaded in under 3 secs’ split by new vs return, for example.

I’ve done quite a bit of work with ‘Google Analytics Page Speed’ reports & have found them really useful, but with plenty of hidden caveats in the data. Feel free to ask any questions if I can help, & do share this with others if you think it would be useful.

Acid Attacks & Online Crime Research

There’s a story in the news at the moment about a woman (Naomi Oni) who was attacked with acid on the way home from work, almost lost her sight as a result, and had her skin badly burned.

The Follow-up Story: “Was it self-inflicted????”

A series of follow-up stories ask “Did she actually do this to herself???“. Here are a couple of headlines from The Independent & The Daily Mail:

inheadline
Headline from the Independent
dmheadline
Headline from the Daily Mail

The Two Facts Behind These Headlines

Each of these headlines is based on 2 facts (the second is key):

  1. The police have taken her laptop as part of the investigation
  2. Naomi Oni had previously searched Google for terms related to ‘acid attacks’, having seen a documentary on a woman called Katie Piper, who was herself attacked with acid.

The detail of exactly which terms she searched for is not included in any of the articles; only that she had searched & visited resulting pages related to acid attacks, and related to Katie Piper.

A Missing Fact: People Search For This Stuff Whenever it Hits Headlines

I don’t know the details of Naomi Oni’s case (the key detail would most likely be the actual terms she searched for), but it’s worth pointing out that whenever an unusual crime is featured on TV/in newspapers, people search for the crime itself in droves online.

Here are a couple of trend graphs to bear that out. Each of these graphs shows the period within which Naomi Oni’s story hit the headlines, and when she herself was featured on TV. As you can see from both graphs, there is a massive spike of interest in searches online when the story hit the headlines, with a minor bump in January, and a much bigger bump when it was featured more widely nationally.

Google Trends Data for ‘Acid Attack’:

 

acidgoogle

Note the bumps in Jan/Feb. The first as the story is picked up, the second as it is featured in greater detail after she gives an interview.

Wikipedia Page View data for the ‘Acid Throwing’ page:

acidthrowing

Again here, a small bump on the wikipedia ‘acid throwing’ page as the story is first covered, and an enormous spike that coincides exactly with the broadcast of an interview with her, covered by the BBC News.

Summary

  • This woman may or may not have inflicted this on herself, but the detail in the articles (the fact that she had searched for acid attacks online having seen a documentary) is just not enough to draw any sort of conclusion at all – it is a very common behaviour.
  • The fact that she searched for ‘acid attack’ having seen a documentary on Katie Piper is not unusual in the slightest.
  • The knock-on effect of this woman’s story is that tens of thousands of other people have now searched for ‘Acid Attacks’ & ‘Acid Throwing’ onliine.
  • Every newspaper mention of ‘acid attacks’, or any unusual crime, essentially acts as an ad for the phenomenon, leading people to further research it online.

The Next Big Challenge for Responsive Design

Here’s a graph showing interest in the phrase ‘Responsive Design’ over the last couple of years:

responsive
(the big dip toward the end is christmas – everyone forgot about responsive design for christmas it seems)

Responsive design is (as you probably know) where you design a website/pages in such a way that the content displays differently on different devices. Usually that means navigation/content is organised differently  on phone vs desktop/laptop (and sometimes differently on tablets, and at other screen sizes).

One of the largest problems for this tactic over the last 6 months has been that – whereas screen resolution (the number of pixels displayed) on smartphones used to be roughly the same across all phones – this now varies wildly. For example:

  • Old iPhone screens were 320 pixels wide by 480 pixels high (320×480).
  • The Samsung Galaxy S3 has a screen resolution of 720×1280 pixels.

In other words, only taking into account those 2 phones, if you’re designing a ‘mobile site’, you’re actually designing for 2 very, very different displays: One that displays 153,600 pixels, and one that displays 921,600 pixels (ie. 6 times as many pixels).

As a result, there are many sites that were designed when 320×480 resolution phones were the norm, which now look very odd when viewed on newer phones.

The New Problem: Laptop screen resolutions are about to become radically different

The new problem is very similar to the phone pixel problem, but across laptops. For the last few years, the majority of laptop sales have been among laptops with roughly similar screen resolutions., give or take 10-20%.

Over the next 6 months however, laptop manufacturers are focussing on much, much higher screen resolutions. Here’s one example to illustrate this:

  • A current 11″ Macbook Air has a resolution of 1366×768 pixels. (ie roughly the same as a Galaxy S3)
  • The Google Chromebook Pixel has a resolution of 2560×1700 pixels. (ie. roughly twice as wide as the macbook air screens, and a massive eight times the width of an old iPhone)

Here are 2 mockups showing what a website will look like on a Macbook Air, and how it will look on the Google Chromebook Pixel.

Keep in mind the 2 devices have roughly similar physical dimensions. In other words, the screen you are looking at will be the same size, but the content displayed on it will look radically different:

What You’ll See on a Google Chromebook Pixel:

econ2

What You See on a Macbook Air:

econ6In other words, both laptops are (very roughly speaking) a similar size, but the content displayed on them will look utterly, radically different in scale.

 The Compound Problem

If you add this new problem to the existing ‘phone resolution’ problem, it means that to simply design your site to look ok on ‘all phones’ and ‘all laptops’, you have to consider a massive range of resolutions, including:

  • Old iPhone: 320×480 pixels.
  • Samsung Galaxy S3: 720×1280 pixels.
  • Current 11″ Macbook Air: 1366×768 pixels.
  • Google Chromebook Pixel: 2560×1700 pixels.

Add in tablets to this, and more forthcoming Apple Retina products, and it becomes very messy firstly to design across all of this, and perhaps more importantly to optimise for conversions/user experience.

Google Plus UI Tweaks

They’ve just changed (quite significantly) the way the ‘+1’ & ‘share’ buttons on Google+ posts look. I thought it would be worth going back and looking at how things looked when it first launched vs how it looks now.

Below are 2 screengrabs of exactly the same Google+ post, 18 months apart. The first image shows how it looked at the time it was posted. The second image shows how it looks after 18 months of subtle, incremental UI tweaks.

Interesting to compare the 2, pick out the differences, and think about why they’ve made the changes.

 

gplusui

Nine Google Analytics Changes

Nine Google Analytics Changes

 
Google Analytics have rolled out a series of fairly big changes to their User Interface. They do this every so often, either to incorporate extra functionality, or simply to clean up the user interface & improve user experience. (See the tweaks they made last time, for example).  This time around most of the changes are purely around the interface, though there are some large, useful functionality tweaks too.

This post covers the 9 main changes made to Google Analytics:

1. Top Navigation Changes

The top navigation (and the overall information architecture) has changed quite considerably. There are still 4 options, but they’ve altered:

Here was the old top nav:

older-top-nav

Here’s the new top nav:

old-top-nav

The main changes there (including the bits you can’t see) are:

  1. The ‘Home’ tab has gone. (this used to house the ‘real time’, ‘dashboards’, ‘intelligence’ reports. To be honest the naming of it didn’t really make sense before, and it was really just a collection of arbitrary reports that didn’t have anywhere else to live).
  2. A ‘home’ icon has been added at the far left. This is totally different to the old ‘home’ tab (which took you to the ‘real time’, ‘dashboards’, ‘intelligence’ reports). It simply takes you back to the main list of all the Google Analytics accounts you have access to.
  3. The accounts dropdown (with new globe logo) has become much wider. (this is more useful than it sounds if you have lots of accounts & profiles, some with long names)
  4. ‘Custom Reporting’ has been renamed’ Customization’. (UK people may dislike the addition of the American Z here!) From what I can see, the content behind the ‘customization’ tab is exactly the same as it was previously at the moment, despite the name change.

 

2. New Pinned Top Nav

A small change, but useful. As you scroll down the page, the top nav remains pinned so that you need not scroll  up to navigate elsewhere. In the screengrab below you can see I’m scrolled halfway down the page, but the top nav is still present:

pinned-top-nav

I thought it may have been nice if they’d included ‘date range’ in that somehow, so that you needn’t scroll up to change that, but really that’s a tiny thing.

 

3. New ‘Recent Profiles’ Dropdown.

As mentioned, the ‘accounts’ dropdown is significantly wider. They’ve also added a ‘recent profiles’ area in the dropdown box (I’ve blurred out my recent profiles here, but hopefully you get the picture):
recent-accounts

That won’t be useful to you at all if you only have one account & one profile, but it’s really useful for web analysts with access to dozens of accounts & profiles, making it quicker to jump back & forth between ones you’ve used recently.

 

4. Left Navigation Changes

The left-hand navigation in ‘Standard Reporting’ has changed significantly. They’ve essentially shoehorned everything from the old ‘Home’ tab in here:

left-nav

I’ve screengrabbed those at the same height. As you can see, on the ‘old’ navigation, you can see quite a lot of the ‘help’ info there. On the ‘new’, it’s all pushed way down.

  1. The left-nav has been split into 2 sections: ‘My Stuff’ & ‘Standard Reports’. (below that, off the bottom of the screengrab, the useful ‘help’ area remains).
  2. The old reports that used to be accessed via the ‘Home’ top tab mostly now sit within the ‘My Stuff’ section there.
  3. ‘Advertising’ has gone from the top level navigation. It’s now been folded within ‘Audience’. (That makes a lot of sense. I do a lot of Google Analytics training & I’d often find myself saying “really ‘advertising’ should be in ‘traffic sources'”. I’d always presumed they pulled it out as a nudge toward AdWords, so it’s good to see they’ve chosen the user-centric option of putting it in the most logical place).
  4. ‘Real Time’ now sits within ‘Standard Reports’. (again, this used to sit within the ‘Home’ area)

On the plus side of that: It’s useful to be able to jump to any report within one area of navigation (the left nav). (Pro tip: You can even jump to customised reports here if you add them in to ‘shortcuts’). Another nice small tweak here is that the date range never resets like it used to when switching between the top navigation tabs.

On the negative side of it: Though the reports are now ordered logically, the items I find people use most frequently (‘Traffic Sources’, ‘Content’, & ‘Conversions’) are toward the bottom of the navigation so a bit more fiddly to navigate around. Alongside that, left nav can now become very, very long. Here’s just how long if you fold all of the options out (click for the full image):

 

all-folded-out

 

None of the options concertina up automatically, so it’s easy to get in a position where you have to scroll up & down the page to get to the report you want. The Google Analytics team have been sensible enough to make sure that all of the ‘top level’ options are visible even on smallish screens, but even at 1024×768, you can see that opening any of the menu items means you need to start scrolling. Note only the ‘Overview’ is available from the Multi-Channel Funnels reports in this 1024×768 screengrab:

dropoff

(Extra tip: Note the other recent change in that screengrab, you can now set a custom ‘lookback’ window in Multichannel funnels. Sadly limited to ’30’ at the moment.)

 

5. New Dashboard Design + Advanced Segments

A minor tweak, but the look of the dashboards has been completely redrawn. Here you can see the ‘widgets’ are now neat white squares on a grey background:

dbdesign

The other big, big change you’ll notice if you look carefully at that screengrab is that you can now apply Advanced Segments to dashboards. That sounds like nothing, but it really, really improves how useful they are. Before if you were looking at a dashboard & wanted to know “what would that look like for iphone visitors?” you’d have to completely rebuild the dashboard, one widget at a time, filtered for iphone users. Now you can simply apply a segment (or multiple segments at once) with a few clicks.

 

6. New Dashboard Layouts.

As well as the redrawn ‘look’ of the dashboards, there are now various different layouts:

dbopts

Again, that sounds like a small thing, but it’s really useful, especially with the ‘Table’ widget which used to totally crunch up with the old ‘3-column only’ layout. Here’s an example of a 2-column layout at 70/30:

 

2-column

 

7. Two Additional Dashboard Widget Types

Dashboards are (as previously) made up of widgets. You can add up to 12 widgets to a dashboard. Previously the widget types were ‘metric’, ‘pie chart’, ‘timeline’ & ‘table’.

Alongside the look & feel changes to dashboards, and the ability to apply Advanced Segments, there are now 2 additional ‘widget types’ in Dashboards: ‘Geomaps’ & ‘Bars’:

new-options

Geomap widgets are exactly what they say – simple geographical maps. By default they show ‘The World’, but you can narrow them down by continent or subcontinent:

europe

The other new widget type ‘Bars’ is quite a sophisticated bar graph tool, offering the ability to pivot data, display it in different orientations, include axis values & titles, limit the number of bars displayed, etc. Here’s an example of the setup & all of the options av威而鋼
ailable:

bar-graph

And here’s an example of the output from that:

baroutput

(A silly example, but hopefully you can get an idea of the types of things you could do with that).

 

8. Lots of odd areas cleaned up.

There are lots of small tweaks & bits of cleaning up that have happened. For example, take a look at the graphing on this multi-channel-funnel report:

mcfgraph

Or the tiny colour tweak from green to orange on the ‘prior date range’ block (pointed out by @mattycurry):

colour-tweak

9. Much Faster.

While hard to quantify, or to get across via a screengrab, the entire interface feels noticeably faster. Part of that seems to be an actual change in how fast reports load the first time you view them, and part of that is the effect of being able to navigate a little quicker around the better organised reports. Google Analytics was already way ahead of most of the enterprise tools in terms of analysis speed, this simply adds to that.

Summary

The changes they’ve made here are fairly large, but not so big as to confuse a regular user for too long. They are essentially improvements to the user interface, with just one or two (very useful) functional additions. The functional tweaks are largely around additional dashboard options. The interface tweaks are largely around information architecture, and improving the speed at which an analyst can make their way around the system.

There are a few downsides to some of the tweaks, but in the main they are really positive & don’t take anything useful away.

Have a play around, and do drop a comment if you think there’s anything interesting not covered here.