Blog 5

Even though the advertising industry is a 200$ billion dollar industry, which is about 2% of GDP, it has been difficult for economists to study and understand it.  However, on-line advertising is contributing to the ability to cross this gap through the possibility of linking advertising to conversions at a low cost and that ads that are served can be served randomly in order to reduce biases.  Although internet data has allowed companies to focus on click through rates, it must be remembered that these are only an intermediate step from the ad being served to the ultimate goal that is desired. However, these variables are used as dependent variables that can be matched with tendencies and intentions.  The data that is now very collectable and provides an opportunity to create algorithms also leads to the potential to draw incorrect assumptions based on the large amount of data.  Most models assume that if one has not clicked on the link, it does not affect purchasing behavior this may not be true because some may complete the purchase off-line.  Also, models believe that if one clicks on a link and makes a purchase, it is all due to that advertisement.  Other models that consider users who have been exposed and those that have not been exposed when generating typical purchase behavior may be inducing bias because the results while factual may correlate with other variables that could be statistically significant but have not been taken into consideration leading to unobserved heterogeneity. It is important to remember that for established brands people might be using click through rates as substitutes for other methods that people would be using to navigate to the site.  Also, it is important to remember for those stores that are brick and mortar, people often use the web to do their research and then will make their purchases at the storefront.  This will induce a negative return on investment when comparing online advertising to online purchasing.  This study indicates that if people are shown an ad or a placebo, people that are shown a placebo are still more likely to take an action based on their browsing activity during the time frame.  Effective time groups for an ad are normally 1-4 weeks, after that the data begins to decay and ads noise to the results.  One of the problems with these studies is small control groups in order to minimize costs. 

On-line advertising has opened up the doors for another method of discrimination.  As part of a screening process for job applicants, potential employers will often search in Google or on another search engine for the name of the applicant.  It seems that for names that are more likely to be associated with someone of African American descent, ads with a negative connotation are showed.  More specifically, it appears that ads that indicate that a person might have had a criminal background are delivered more often for names that would be associated with someone of African American descent than with a name that would typically be associated with a Caucasian.  Names were selected that were thought to be associated with African Americans (both sexes) and Caucasians (both sexes).  In order to validate the name selection, the names were searched and then images that were associated with each name were viewed.  The selected examples were then searched to see which advertisements were served, it was indeed more likely that those indicating an arrest showed for names associated with African Americans.  How is this selection done, who dictates that the ads show up for certain “racially associated names”? As compared to text ads which are seen and viewed the same by everyone, on-line ads can target based on history and indicative information that is provided by the user.  Online discrimination can have negative implications which people will complain about.  However, are also going to view media that is enjoyed and targeted to them based on the same parameters as discrimination or will they be thankful for it.  Unfortunately, these ads are probably based on a combination of statistics and intuition.  While it may not be acceptable to target people with ads that have negative connotation, by restricting what can be targeted or used as key words, half of the benefit of using on-line marketing is eliminated.  Discrimination is a slippery slope; targeting older people with certain ads could be age discrimination.  I think that one of the best things to do would be to educate people on the capabilities of on-line marketing and make them aware of such practices.  

The Google Analytics Fundamentals course was much more time consuming than it was supposed to be.  However, in order to watch all of the videos and complete all the tests it needed to be.  It was very nice though that the Google Analytics Platform Principles did not take nearly as long to complete and that it did not go into as much detail.  It was a good overview that helped sum up and simplify all of the details that were exposed in the first assignment. 

In Google Analytics, you are able to filter out the data that is contributed from employees, that way the data is not skewed.  Google Analytics also enables you to import data from multiple Google interfaces and other non-Google sources.  It also provides the ability to load it directly to third part reporting systems. However, reported data is only as good as the collection, configuration, and processing allow it to be.  Website users are tracked in a hierarchy of visitors, sessions, and interactions.  Processing and configuration work in conjunction with each other to make data valuable.   One of the most important things about Google Analytics is that once a filter is applied and data is collected, the data that is excluded cannot be pulled in for analysis.  If our client were able to apply Google Analytics, we would have to include the tracking information on every page in her web page that we wished to track.  If we did not, it would not record and store the data for those pages.  However, right now as our client has not successfully been able to turn on Google Analytics, this will not do much good for us right now.  Being that I work with data and large data tables every day and know how to filter or use pivot tables to analyze my data, these lessons have taught me that it is better to collect data for everything and worry about sorting it later.  It is also important to remember that results are only as good as the data that is collected and used, it is essential to know that the data you are collecting and using to make results is truly indicative of the process.