Statistics for Research

1
Statistics for Research

• Spring 2004

2
Lies and Statistics

• Statistical Lies
• 14 April 1999
• Dan Knight
• Mark Twain is purported to have said, "There are
  lies, damned lies, and statistics."
• Whether he actually said it or not, the fact
  remains that statistics can and do lie. Of all
  computer users, Mac users should know it best.
• Remember the 3 and 4 market share figures we
  saw when Apple seemed on its deathbed? Well, they
  were only partly true -- they only measured part
  of the computer market.
• Worse, market share numbers only apply to current
  sales. They tell us nothing about the installed
  base. So even when Apple had a 5 market share,
  anywhere from 10-15 of all computer users were
  using Macs. (See Mac installed base, MacInTouch)
• After all, we don't sell our existing computers
  every month, quarter, or year!

• More Lies
• A recent study by WebSide Story showed that only
  3 of web users are on Macs. This is suspiciously
  low for a platform with at least 10 of the
  installed base!
• The problem isn't the numbers, but how they were
  derived. WebSide Story used HitBOX Tracker. As
  Ben Wilson of MacCentral notes, "It should be
  emphasized that the HitBOX service only provides
  a viewer for Windows, meaning not many
  Mac-specific sites are likely to use the
  service."
• Simply stated if you use a platform related
  measuring tool, results will be skewed toward
  that platform.

3

• Real Statistics
• A survey of visitors to Low End Mac shows very
  different results from those created by WebSide
  Story
• 58.5 use the Mac OS (not 2.7), and only 36.6
  use Windows (not 94.5)
• 63.8 browse with Netscape, 34.3 with Internet
  Explorer (not 68.7), and 0.7 use iCab
• Admittedly, these numbers show a bias toward the
  Mac side, just as the WebSide statistics show a
  pro-Microsoft bias in both OS and browser.
• The Truth
• The truth is, we'll never have precise figures
  comparing Mac OS, Linux, Windows, and other
  market share or installed base. And any survey of
  the internet will be biased by the sites and
  software used -- sometimes just a little, but
  sometimes a great deal.
• In reality, Mac users comprise somewhere over 10
  of the installed base, but probably less than
  15. Based on that and the ease of connecting all
  but the oldest Macs to the internet, we can
  estimate that Mac users are at least as likely to
  be on the internet as Windows users, so at least
  10 (and perhaps as high as 20) of internet
  users are Mac users.

• Anyone purporting to offer statistics on
  something as vague as the installed base or use
  of the internet with three or four digits of
  precision is being unrealistic. While the numbers
  may mathematically provide a figure of great
  precision, the nature of the study leaves a
  margin of error of several percentage points --
  just like those public opinion polls in the press
  and on TV.
• We can safely say that Windows users outnumber
  Mac users in the home, in the office, and on the
  internet. We can estimate there are about seven
  times as many Windows users as Mac users.
• Beyond that, we have to remember that there is a
  great margin of error in these surveys. I would
  be a fool to extrapolate from visitors to my site
  and claim that almost 60 of all computers on the
  internet are Macs, despite the numbers which
  could be construed to "prove" it.

4
Another Example

• TECHNOLOGY BREAKDOWN Numbers Don't Lie - Or Do
  They?
• Statistics are a funny thing. You can use them to
  prove almost any argument. For example, many
  people regard Philadelphia 76er's player Allen
  Iverson as a great player. To be sure, while I
  despise AI's off court persona, there is no
  denying his prowess on the basketball court. One
  would expect that such a talented player would be
  able to carry his team like Michael Jordan did
  the Chicago Bulls. However, in case of the 76ers,
  the converse is true. Statistics show that in
  games where Iverson scores 40 or more points, the
  76ers actually have a losing record. The same was
  true of another 76er great, Wilt Chamberlain.
  What makes the use of statistics persuasive then
  is something called correlation. Simply put,
  correlation is what defines cause and effect
  between two observable phenomena - in this case,
  the number of times AI scores 40 more and the how
  many times Philly wins when he does. In our
  example, the level of correlation is pretty high.
  However, when the correlation between events is
  low, you can make pretty much any argument you
  want - statistics then becomes the smoke and
  mirrors used by spin doctors to shape the views
  and opinions of those who chose to, as the Wizard
  of Oz said, "ignore that man behind the curtain!"
  Right now, Republicans are pretty happy with
  themselves. On Sunday (December 14, 2003) US Army
  forces captured Saddam Hussein. But more
  importantly, the Bush Administration has been
  pushing the notion that the economy is
  recovering. Well, let's take a look at some
  numbers. The administration says unemployment is
  down. But what numbers do they use to support
  that claim. The most quoted statistic with regard
  to unemployment is the number of new claims for
  unemployment aid. If that number goes down,
  surely the unemployment problem is getting better
  right? Not necessarily so. In November, the
  government stated unemployment rate fell from 6
  percent to 5.9 percent and is the lowest since
  the 5.8 measurement in March 2003. What could be
  wrong with that? It turns out plenty.

5

• First, the statistics are based on those people
  who are unemployed and are collecting
  unemployment benefits. One of the biggest
  weaknesses in the government's numbers is that
  they do not track what happens to people who
  exhaust their unemployment benefits and fall out
  of the system. Second, fluctuations in the
  unemployment rate give no indication how long
  people remain unemployed. Of the 8.7 million
  people classified as unemployed, nearly a quarter
  of them (23.7 percent) have been unemployed for
  over six months - which turns out is the highest
  level of long term unemployment for over 20
  years. Finally, let's not forget a statistic that
  is one of my favorites. Over the last 10 years,
  the unemployment rate in the African American and
  Latino communities has tracked at roughly double
  the national average. Apparently, Mr. Clinton's
  economic policies were about as beneficial for
  American minorities as Mr. Bush's - not very. In
  the name of fairness, Democrats like to take
  credit for the prosperity of the late 90's. The
  only thing that Mr. Clinton and his bunch had to
  do with the boom of the 90's was that they were
  sitting in the chairs at the time. The boom of
  the late 90's was due primarily to the emergence
  of the World Wide Web and a "gold rush" of
  speculative investment that followed and
  eventually collapsed. In fact, Clinton era
  legislation such as the 1996 Telecommunications
  Reform Act probably did more to stifle
  competition by shifting the balance of power in
  favor of the Bells (local phone companies) as
  opposed to long distance providers and
  competitive local exchange carriers (CLECs or
  alternative phone companies). That, and we won't
  even get into what the Telecom Act did to the
  media business. So where is the economic
  recovery that we keep hearing about? It's in the
  numbers and how you look at them. What we've
  shown here, or at least opened the door to is the
  possibility that economic recovery is not very
  tightly correlated to job creation and low
  unemployment. We can say this because
  productivity is up while long term unemployment
  is high. The reason for this is due to a number
  of factors, but most notably automation. One of
  the primary factors that allowed the
  manufacturing industry to move overseas was that
  automation got affordable. Automation means equal
  or increased productivity with fewer employees.
  This translates to lower costs, which in turn
  translates to a stronger cash position which
  helps corporate management keep company stock
  prices (and their bonuses) up. To bring this
  point home, Lehman Brothers economist Drew Mathus
  commented that reported 2 increase in the number
  of aggregate hours worked in the economy was "the
  equivalent of adding 350,000 additional jobs."

6

• On Tuesday, the Department of Commerce issued a
  report of the "Digital Economy" that concludes
  that signs of recovery are present in the
  technology industry. Again, a look behind the
  numbers tells a different story. For instance,
  the report indicates that the rate of change in
  IT employment decreased by 9.2 percent from 2001
  to 2002, six times faster than the rate for all
  other private industries combined which lost jobs
  at a rate of 1.6 over the same period. The report
  also notes that the average wage for IT workers
  decreased by 1.3 percent from 2002 - 2003 while
  that of the average worker increased 1 percent
  over the same period. Call me crazy, but that
  doesn't sound like recovery to me. A few months
  ago, I wrote a column that talked about the dim
  prospects for the future of American IT
  employment, mostly due to offshoring and foreign
  outsourcing. The Digital Economy report states
  that of the 516,000 IT jobs lost since 2000,
  nearly half have been highly skilled positions,
  you know engineers, programmers and the like. To
  avoid ticking off potential campaign
  contributors, the government report draws no
  conclusions of its own, but reiterates those of
  other researchers. One of the conclusions drawn
  by Goldman Sachs and cited in the report was that
  "job losses due to offshoring could reach 6
  million over the next decade." More frightening
  was Goldman Sachs' conclusion that "In the end,
  increased offshoring would reduce labor demand,
  raise imports, and place downward pressure on the
  value of the dollar." The long and short of it
  is, that those on both sides of the political
  aisle can use statistics to prove whatever story
  is convenient at the time. Next year, we will
  once again be faced with choosing a new
  President. When some campaign flunky starts
  telling you about how great their person is, and
  offers a few statistics, take a look behind the
  numbers because as we now know, numbers can lie.
  Russell de Pina is a Principal with n2active, a
  technology consulting firm located in Long Beach,
  CA and Houston, TX. Russell can be reached by
  email at rdepina_at_n2active.com and
  eurfeedback_at_eurweb.com

7
One final Example

• Lies, Damn Lies, and Statistics    
  Publishing Free or Fee?
• BY Vin Crosbie August 13, 2002
• Mark Twain popularized in the U.S. Benjamin
  Disraeli's statement, "There are three kinds of
  lies lies, damn lies, and statistics." That
  sardonic journalist, author, and speaker
  recognized the persuasive power of
  authoritatively made numeric presentations to a
  largely innumerate public. Ninety-two years after
  Twain's death, the Online Publishers Association
  issued a statistical report that buoys advocates
  of charging for online content.
• Entitled "Online Paid Content U.S. Market
  Spending Report" and prepared by survey firm
  comScore Networks, the report states U.S.
  consumers spent 675 million for online content
  in 2001, a 92 percent increase over the previous
  calendar year.
• Advocates of charging for online content point to
  this report's statistics as a signal consumer are
  more willing to pay. The statistics have induced
  journalists into affirming a trend is underway.
  "A Shift Registers in Willingness to Pay for
  Internet Content," headlines The New York Times.
  "Americans are warming up to paying for content
  on the Web," adds CNET's News.Com.
• Wow! Consumers are paying for online content.
  What is it that made them suddenly more willing
  to do this?

8

• To answer that question, I must quote another
  famous American, William Jefferson Clinton "It
  depends upon what the meaning of the word 'is'
  is." That presidential statement to a grand jury
  is now enshrined in Bartlett's Familiar
  Quotations, which is edited by Justin Kaplan,
  Mark Twain's biographer.
• Clinton's quote is an exercise in semantics. So,
  unfortunately, is the OPA's report.
• OPA is "an industry trade organization dedicated
  to representing high-quality online publishers
  before the advertising community, the press, the
  government, and the public," whose membership
  includes Bankrate.com, CBS MarketWatch, CNET
  Networks, CondéNet, Cox Enterprises, ESPN.com,
  Forbes.com, Knight Ridder Digital, Le Monde
  Interactif, MSNBC.com, New York Times Digital,
  Salon Media Group, Scripps Network, Slate,
  SPACE.com, Tribune Interactive, USATODAY.com,
  Wall Street Journal Online, Washingtonpost.com,
  Newsweek Interactive, and weather.com.
• Those companies are publishers and broadcasters
  of what most people traditionally term content
  news and features, or TV programming. OPA
  members' parent companies would likely define
  content that way.
• How does the OPA report define paid content? "We
  restrict our definition of 'paid content' to
  digital intellectual property purchased through a
  Web browser by an individual."
• Paid "digital intellectual property." What does
  that mean exactly?
• To OPA, it means not only what its membership's
  parent companies would traditionally define as
  content but also business-to-business (B2B)
  online research, such as Internet industry
  reports purchased from eMarketer.com wiring
  reports from the Institute of Electrical and
  Electronics Engineers' site day trader
  investment advice (e.g., ChangeWave.com) and
  downloadable clip art (e.g., ArtToday). All this
  B2B material counts toward consumer paid content
  totals.
• Also included is pure research, such as
  Merriam-Webster OnLine, eLibrary.com,
  Brittanica.com, and USSEARCH, a category
  including not only dictionaries and encyclopedias
  but also public information and personal
  background searches. To OPA, content also means
  online subscriptions to services. OPA counts
  "Community Directories" such as Ancestry.com and
  Classmates.com. The services OPA counts as
  content include online consumer credit help such
  as ConsumerInfo.com and CreditExpert, plus
  "Personal Growth'' services, such as eDiets and
  WeightWatchers.com.

9

• OPA includes online greeting cards services
  (AmericanGreeting.com, Blue Mountain Arts, and
  Hallmark.com) as content. Oh, gift certificates
  purchased from those sites are content, too.
• OPA says "Entertainment/Lifestyle" streaming
  media from Real.com and pressplay and even pinups
  from Playboy.com are content. The study further
  counts online game site subscriptions, such as
  the Alien Adoption Agency, Case's Ladder, and The
  Well Dressed SIM.
• How far does OPA's definition of content go? Toss
  in personal ads and dating site subscriptions,
  such as Match.com, Singles.com, and kiss.com.
• "Digital intellectual property"? I can understand
  Brittanica.com's content being defined that way.
  But a personal ad labeled "Sexy Guy Seeks Comfy
  Girl" on kiss.com? A post of "Uma's
  Unmentionables" on The Well Dressed SIM? Gift
  certificates on Hallmark.com?
• Do I detect OPA inflating the definition of paid
  online content? Why not add online revenue from
  Ticketmaster and other traditional services?
• At least OPA decided to "exclude software
  downloads, pornographic sites, gambling sites,
  and certain other classes of sites which, in our
  view, skirted the bounds of decency or the law."
  Interesting. You might think software would fit
  the description of digital intellectual property.
  I wonder what bounds of decency or law thousands
  of companies that offer downloadable software are
  skirting? I can see excluding gambling
  subscriptions and winnings as a form of content.
  Incongruous that OPA would exclude pornographic
  downloads, perhaps the most widespread form of
  online paid content.
• Does OPA's definition of online paid content make
  sense when you look at its membership? No, but it
  sure fabricates a great amount of money OPA can
  claim is being paid for online content.
• Let's look again at the 675 million the report
  says was spent on online content in 2001. Here's
  the breakdown

10

• Business Content -- 214.3 million
• Entertainment/Lifestyle -- 112.0 million
• Personals/Dating -- 72.0 million
• Research -- 57.9 million
• General News -- 51.8 million
• Games -- 46.5 million
• Community Directories -- 46.1 million
• Credit Help -- 32.4 million
• Personal Growth -- 24.7 million
• Sports -- 10.0 million
• Greeting Cards -- 6.8 million
• Cut categories that actually provide paid
  services (Personals/Dating, Community
  Directories, Personal Growth, and Greeting Cards)
  and 675 million drops to 523.4 million. Then,
  eliminate downloadable songs and pinups. The
  figure arguably falls to 411.4 million. Remove
  B2B content from the Business Content and
  Research portions of this consumer spending
  report, and the total spent online for content,
  as traditionally defined, was probably only about
  139.2 million last year.
• Indeed, the only OPA members listed in the
  report's ranking of the top 25 paid online
  content sites were The Wall Street Journal Online
  and ESPN.com.
• Did American consumers spend 675 million for
  online content last year? Depends how you define
  content. Most people wouldn't define it the way
  OPA does. But OPA's definition sure makes for a
  spectacle of a press release.
• "If a spectacle is going to be particularly
  imposing, I prefer to see it through somebody
  else's eyes, because that man will always
  exaggerate. Then, I can exaggerate his
  exaggeration, and my account of the thing will be
  the most impressive," Mark Twain also said.
• It seems the wishful revenue hype of the Internet
  bubble era isn't entirely over. It's doubly
  ironic when you consider traditional media
  outlets belonging to OPA's membership are
  headlining the business stories about inflated
  revenues.
• OPA states, "This first report and its ongoing
  installments will enliven the debate about the
  role of paid content in online publishing." It
  did for me. There is a role for reporting the
  amount of paid content in online publishing, but
  there's no role for hype.

11

• Which of the two types of pasture (danthonia
  sub/clover, phalaris sub/clover) has a faster
  growth rate between July and October (in
  kg/ha/day)?

12
(No Transcript)
13
(No Transcript)
14
Class Objectives

• Understand the scientific rationale behind
  different statistical methods.
• Understand the socio-political implications of
  the use of different statistical methods.
• Understand the different approaches to statistic
  research methodologies descriptive and
  inferential methods as they apply in the natural
  and social sciences.
• Critically analyze those statistical methods
• Discuss the Ethical dimensions of statistical
  research.
• Understand the planning and preparation stages
  of statistical research through its application.
• Apply the skills and theories discussed within
  the context of the Prescott and Las Vegas areas.
• Explore how well structured and rigorously
  acquired knowledge on social and environmental
  problems can be a powerful tool in finding their
  solution..

15
Course Outline

• I.  Introductory Concepts
• A.  Why are we here If statistics are tricky,
  why should we learn about them?
• B.   Types of data
• C.    The Basics of SPSS
• D.    Populations and sampling
• II. Basics of Descriptive Statistics
• A. The Standard Normal Distribution
• B. Descriptive statistics of location and
  dispersion
• C. The Binomial and other Distributions
• III. Inferential Statistics The use of
  statistics to test hypotheses and establish
  connections and causalities.
• A. Comparing descriptive statistics
• B. Analysis of Variance (ANOVA)
• E.  Correlation
• F.  Chi-Square and non-parametric tests
• G.  Simple and multiple regressions
• H.  Evaluating inference and manipulating data to
  meet the assumptions of statistical models.

16
Evaluation

• Contract
• 4 Computer Assignments
• Take home exam in Vegas and Prescott

17
Text

• Kirkpatrick, L. and Feeney, B. (2001) A Simple
  Guide to SPSS for Windows. Belmont, CA,
  Wadsworth. 118 pp.

18
Course Scheule