This post is a continuation of my earlier posts (here and here) on using Google Insights to look into Google’s data on searches for Canadian political parties and leaders during this election season. Although I have many qualms about the use of this data and how generalizable it is, I am so far impressed with how well it syncs with the general zeitgeist about the ebb and flow of party popularity (although I havn’t figured out how to quantify that relationship yet — possibly by indexing the data from Google with polling data?). Obviously the population of Canadians using the internet — specifically Google — to search for political parties and leaders is not exactly the same as the voting population, and searches are not in themselves an indication of which party’s candidate someone is going to vote for. I do think it is interesting, however, how the search data correlates with some general shifts during this election, including the bump the green party got during the debate controversy and now the rise of the NDP as a serious contender, or at least a possible Liberal spoiler, in this election.

Today’s data follows the same format and configuration as my other two post:

The Set Up

The Search Terms and Filters are very important when using Google Insight. As I found in my initial post “Searches for Canadian Political Parties and Leaders: Google Insights Data” it was very important to include specific search terms people might use, while excluding search terms which brought in extraneous data. The search terms I constructed were based on ballancing these two concerns. As before, I am including data from all of Canada’s regions and from the past 30 days.

The Interest Over Time graph:

Interest Over Time

A couple things of interest here. One is that the green party post-debate controversy bump has continued, and the total frequency of searches for the green party and their leader have only fallen bellow searches for the Liberal Party and their leader on three days. More interesting, I think, is that since September 17th the NDP and Jack Layton have been in the Canadian political search rave. While this may relate to how connected Canadians and Google user’s skew politically it should be noted that before the 17th the Conservative and Liberal parties were consistently number 1 and 2 in terms of search frequency.

Now the regional interest graph, this time focused on the NDP:

Regional Interest Graph

In general I think Google’s data here seems to reinforce other things I have been heading. The NDP seems strongest, however, in provinces where the Green party does best. I think it would be more interesting to focus on individual provinces like British Columbia and see if Google’s data for is deep enough to actually look into city by city search trends. If it is deep enough to allow such analysis, at least for populated provinces, it could be used to attempt to predict which way ridings might go on election day.

Finally, the same caviets that I wrote in my first post apply here as well.

Wordle is a new visualization toy that creates word clouds from blocks of texts (blog posts, books, stories, speeches, etc.) in which the size of a word represents that word’s relative frequency compared to other words in the text. I’ve been playing with Wordle for the last couple days, mostly using texts from the history of psychology (old journal articles) as preparation for a possible future thesis project. Yesterday I decided to use the opening election speeches from three of Canada’s national political parties, the Conservative Party of Canada, the Liberal party of Canada, and the New Democrat Party. I have not included anything from the Green Party of Canada or the Bloc Quebecois because I could not find comparitive material.

Stephen Harper, Sept. 7. 2008 – Election a choice between certainty and risk:

Stephane Dion, Sept. 7, 2008 – A new path for Canada:

Jack Layton, Sept. 7, 2008 – Its time to choose change:

Conclusions:

• I do not pretend that these kind of graphs meen anything conclusive.
• Its interesting how prominant the words “promised” and “delivered” are in Harper’s speech.
• All three speeches use variations of Canada and Canadians extensively.
• Jack Layton’s speech seems to be chanelling Obama a little, with prominant use of the words “change” “believe” and “can”. It would be interesting to do a comparitive word cloud and see just how close they really are.
• That said, Harper’s speech seems to be better ballanced, word frequency wise, and contain more “issue” words than either the NDP or Liberal party speeches.

The Canadian Green Party seemed to have enjoyed a significant bump in it’s online (and offline) profile during the debate controversy, so I figured it would be interesting to take another look at the Google Insights’ data and see what effect this has had on internet searches for political parties and leaders. As you can see bellow, the (now reversed) decision to exclude Green Party leader Elizabeth May from the leader’s debate resulted in a significant surge in searches for the party and it’s leader.

First, the set up, using the same settings as my previous Google Insights search:

The key:

The 30-day Interest Over Time graph:

As you can see, according to Google’s data, searches for the Green Party, et. al. surged along with the other parties between September 6th and 8th, and then surpassed the other parties between the 8th (when the decision to exclude the Green Party from the Debate was announced) and 9th and as of September 10 they remain in first place. The afternoon of the 10th it was announced that Elizabeth May will be allowed to participate in the debate.

Finally, the Regional Interest graph, focused on the Green Party and Elizabeth May:

Thoughts:

• Searches for the Green Party of Canada and Elizabeth May continue to be strongest in Nova Scotia, Ontario, Manitoba, and to a lesser degree, British Columbia.
• I am not sure why Saskatchewan is not showing any search data right now, it may be because Google has not yet collected all the data yet. I’ve noticed a two to three day delay on Insight data.
• It seems that the regions with the strongest Green party searches are also home to the strongest NDP searches, although I have not figured out a way to quantify that data yet and test it.
• It will be interesting to see how quickly the Green Party looses its new first place status (assuming it does) as the election moves forward.
• That said, the same caviets i had about my orginal Insights search apply here as well.

Google Insights is an interesting set of tools that allow you to analyise the relative popularity of search terms over time and by geographical regions. To quote Google:

Google Insights for Search analyzes a portion of worldwide Google web searches from all Google domains to compute how many searches have been done for the terms you’ve entered, relative to the total number of searches done on Google over time. We then show you a graph with the results, indicating interest over time, plotted on a scale from 0 to 100; the totals are indicated next to bars by the search terms.

I first became interested in using Google Insights in order to track the spread of opposition to Bill C-51 and the influence of certain players in that particular debate. While I will probably do some tracking of scientific and medical memes in the future, yesterday I decided that in the spirit of the election call here in Canada that I would look into political searches.

Combined Party Names and Leaders Search

The first step is to create the terms terms that Insights will use. This step is tricky and requires allot of playing around and testing to get right. I wrote this entire post using search terms that I later realized were inappropriate because they were pulling in too many unrelated searches. The key seems to be to only use search terms that are almost exclusively used to refer to your topic of interest, while excluding terms that are used for unrelated searches.

For example, Canada’s “New Democratic Party” is frequently refered to as the “NDP” (in fact, the search term “New Democratic Party” itself has very little profile on Google Insights), yet nobody really calls the “Liberal Party of Canada” the “LPC”. Acronyms are probably especially dangerous because they are frequently used to refer to multiple unrelated things, so it seemed best to avoid them in generally. However, I make an exception for the NDP because of the previously mentioned tendency and in order to fill out the searches.

As you may have noticed I have excluded the bloc quebecois party and only include national parties in this search. This is not because of any personal political beliefs, but because of problems with translation and search term selection which I did not think I had the knowledge to properly construct. I have included the bloc quebecois’s leader Gilles Duceppe in the Leader’s Names seach bellow.

Next you set the filters that are applied to the analysis. In this case I have limited the the Search to Canada, including all subregions (Google uses the provinces to break Canada down into subregions) and set the date range (for now) to the last 12 months. I have left the Categories filter to “All Categories”.

Interest over Time

The first graph presents “Interest Over Time” over the 12 months ending September 7th:

Just to be clear about what the numbers represent, I will quote Google’s explanation:

The numbers on the graph reflect how many searches have been done for a particular term, relative to the total number of searches done on Google over time. They don’t represent absolute search volume numbers, because the data is normalized and presented on a scale from 0-100; each point on the graph is divided by the highest point, or 100. The numbers next to the search terms above the graph are summaries, or totals.

Here is another “Interest Over Time” graph, this time using the same search terms but including data from only the last 30 days:

Regional Interest

There is also separate graphs for each region (province) of Canada and for each search term. Here is the results for the 12 month search, as the 30 day search has too little information to include all parties and provinces. Each graph is done by search term and includes a color coded map of Canada.

Conservative Party:

Liberal Party:

The New Democratic Party:

Green Party:

Top Related Searches

Insights allows you to export the search data to a CVS file which includes information not directly availble through their graphical interface. This includes an expanded list of related searches for each search term. This could be useful as a way to test how appropriate your search terms are. That is, looking through the related search terms can help you make sure that the search terms you used was narrow enough to exclude completely unrelated searches. For example, in an earlier search set I used the term ["Liberal Party" + Dion] and found that, as I should have guessed, the majority of “Dion” searches were looking for Celine Dion, not Stephane Dion.

Furthermore, it is potentially interesting in and of itself to see what people who search for the NDP, for example, also search for. As Google explains:

Top searches refers to search terms with the most significant level of interest. These terms are related to the term you’ve entered; if you didn’t enter in a search term, the top searches will be related to the category or country/territory you’ve chosen. Our system determines relativity by examining searches that have been conducted by a large group of users preceding the search term you’ve entered, as well as after.

Although I have not pursued this type of analysis here, I have included graphs of the relative popularily to the top 10 related seach terms for each search terms I used, using Open Office to construct basic bar graphs.

Top Searches related to ["conservative party"+conservatives+"stephen harper"]

Top Searches related to ["liberal party"+liberals+"stephane dion"]

Top Searches related to ["new democratic party"+ndp+"jack layton"]

Top Searches related to ["green party"+"elizabeth may"]

Party Leaders Search

The following is another Insights search, this time using the proper names of the party leaders as search terms. The terms are enclosed with quotation marks, and thus only represent searches done for full names, that is, “Jack Layton” but “Layton” alone. I did things this way to make things ‘fair’ and exclude other references, for example, if I had included “harper” as a search term along with “Stephen Harper” the results would have included all searchs for “Ben Harper” which I do not think either of them would approve of. Similarily, A search including “Dion” would have pulled in searches for Celine Dion, thus artificially inflating the search data.

Here is the 12 month Interest over Time graph, showing the search “Stephen Harper” having a clear advantage over the other leader’s names:

Perhaps more interesting, here is the 30 day Interest Over Time graph for the leader’s names:

It is interesting that when full leader’s names are used Stephen Harper is has a clear Google Search advantage over the other party leaders. The 12 month Regional Interest graph reinforces this result, at least according to the way I contructed this search.

Some caveats:

1. Google Insights only tracks searches done using Google, and clearly this is a different demographic than the voting population itself.
2. These particular searches only track the terms I choose to use, which is definitely not equivelent to all the possible search terms used to search for either political parties of leader’s names. For example, people who wanted to find information about Stephen Harper but searched for “Prime Minister” are excluded from this analysis, as are people who searched for “Harper” alone or “Canadian Political Parties”.
3. I do not believe there is some kind of one-to-one correspondence between what people search for and who they are going to vote for, that said, I find the clear differences in searches for party leaders particularily interesting.

Thoughts:

1. I wonder if, and how, the parties themselves are using this kind of data?
2. Would this kind of analysis be more useful for provincial elections? The data could be restricted to a single province and thus the results should include less unrelated searchers.
3. How could Insights be used to track memes, especially the rise and fall of medical beliefs?