Teaching with Stata

1
Teaching with Stata

• Alan C. Acock
• Peter Lachenbruch
• Oregon State University
• alan.acock_at_oregonstate.edu
• peter.lachenbruch_at_oregonstate.edu

2
Teaching with Stata in a SPSS College

• When I started teaching with Stata
• 90 of the faculty used SPSS
• 10 used SAS
• Stat Transfer is a powerful argument
• SPSS can save a Stata file for people use to
  using SPSS for data management
• Faculty using SPSS rely mostly on its excellent
  menu system
• Students who know Stata can use SPSS menu

2
3
Need Colleagues

• College had good fortune of hiring Tony
• Impeccable statistical qualifications
• Long history of using Stata
• Use Statas strengths to get colleague support
• Long Freese Limited Dependent Variables
• gllamm
• Complex sample capabilities
• ice, mim
• Many of these are user written

4
Need Colleagues

• College Methodology Core presentations
• Students presentations doing things not possible
  with SPSS
• Recruit Faculty Projects to using Stata have
  Faculty presentations
• Emphasize what Stata can do rather than attacking
  SPSS--advantages are obvious
• Faculty inertia is like herding cats

5
Its Hard to Convince Menu Addicts

• Most colleagues have a research team need to
  coordinate their work
• Need to plan data management analysis
• Need to maintain a record of all changes to data
• Need to have comments
• Need to organize understand datasets
• Need to replicate data modifications
• Need to replicate results
• Need to do revise resubmit re-analysis

6
Convincing Menu Addicts

• Appropriate use of menu system
• Nobody ever needs to replicate your results
• No other people working as part of a team
• All your manuscripts are published without need
  for revisions
• Learning about options w/o reading the manuals
• I use them with graphs then copy to a do-file

7
Convincing Menu Addicts

• Refuse to help student w/o all relevant do-files
• I should be able to stick their flash drive in my
  PC and have their data created and analyzed
• Require appropriate comments
• Students never imagine how much they can forget
• Need write comments so somebody else can use the
  program

8
Command Structure--Statas Great Strength

• I show the Stata program in class
• I require the Stata program to be submitted with
  all labs
• In presentations, I sometimes show SPSS or SAS
  code that can duplicate Stata
• without stressing how bizarre it is
• type a little, get a little is compelling by
  comparison to SPSS/SAS
• regress y x1 x2, beta
• logit y x1 x2, or
• poisson y x1 x2
• factor x1 - x12, pcf

9
Working with Graphs

• Many introductory statistics courses minimize
  graphic presentations
• Many students colleagues rely on Excel for
  graphs
• Stata 10 has a professional quality graphs
• Graphics editor in Stata 10 gives students a
  sense of efficacy it is hard to get with most
  statistics procedures
• Stress knowing their audience--bar chart or
  histogram may be most effective for non-technical
  audience
• Stress need to have strong statistical evidence
  behind the graph, even when not reported
• Two-way graphs with multiple overlays--scatter,
  linear, quadratic

10
What to do in Labs--Motivations

• My undergraduates are in the social
  behavioral/health sciences
• All state programs in Oregon are required to
  justify their existence on a regular basis
• Funding for new programs is part of many
  positions
• Substantive content gets first job ability to
  write/evaluate program proposals get promotions
• Labs focus on statistics/graphs, data access,
  analysis that facilitate evaluating programs

10
11
What to do in Labs--Activities

• Class/group conduct a survey near start of course
• N of 100 is nice, 50 is critical
• Let class/group pick the topic
• Each enters some data then use append
• Label variables, values, missing values
• Create a scale using rowmean/rowtotal
• Cover alpha after corr
• Use their survey data to write short advocacy
• Use their survey for several types of analysis

11
12
What to do in Labs--Activities

• Class/group survey
• Require them to have some items that are not
  simple to code
• Skips
• Select all that apply
• Ranking of five items
• This helps them with coding, data entry, data
  management tasks
• Lab reports should emphasize interpretation
  rather than just getting the right number.

12
13
What to do in Labs--UCLA Stata Portal

• Give them assignments to find instruction on
  selected topics, for example
• http//www.gseis.ucla.edu/courses/ed230a2/notes/ef
  fect.html for effect size
• Search Statas FAQ for some topic we provide
• Do exercise based on UCLA video
• Students played jeopardy as review for tests
• Elements of research methods introduced wherever
  I canThis helps to make statistics seem more
  relevant

13
14
What to do in Labs--Activities

• Use real data for most analyses
• Require some data management for many of the
  assignments
• W/O this, they quickly forget how to manage data
• W/O this, they dont understand importance of
  data management
• Require a do-file for many of the assignments
• Regular use of display for calculations gets them
  used to keeping Stata open

14
15
What to do in Labs--Activities

• Have them keep a portfolio
• Include all lab activities
• Include notes from lecture
• Include exercises they did that were not required
• Include activities for class
• I give them full credit if this looks fairly
  complete. It is a small portion of the grade
• They often go back to these in subsequent
  coursework

15
16
What User Written Commands?

• Introductory graduate statistics can utilize user
  written commands
• Here are some I like them to have in first year
  course
• fre
• Those who have SPSS background love this
• ice, mim, misschk
• Missing values are a profound problem in
  social/behavioral/health sciences

16
17
What User Written Commands?

• Here are a couple more user written commands
• Long Freese commands for categorical count
  outcomes
• corrtab
• Nicer display of correlations including
  significance
• I use this when searching for examples of small,
  medium, large correlations for lecture use
• pathreg
• This does path analysis quite simply insures
  each equation has same N.

17
18
What User Written Commands?

• More user commands I include for data management
• Center
• Simple, but useful
• User written commands like this are a great way
  to introduce them to writing their own commands
• zscore
• revrs
• Handy for reversing scores on ordinal variables

18
19
What User Written Commands?

• User written commands I use as teaching tools to
  help students visualize. Many students find a
  visual approach more accessible than a
  mathematical approach
• beamplot mpg, over(rep78)
• Balance mean for interval over categorical
  variables
• clt
• Nice demo of Central Limit Theorem
• chidemo df p for df 1, 5, 30, 50 reminds them
  of the CLT
• cordemo
• Does r .5 mean the relationship is strong?

19
20
What User Written Commands?

• User written commands I use as teaching tools
• pwrdemo2, n(x) diff(x) alpha(x)
• Where diff is the effect size
• Helps students understand effect size concept
  that is widely used in psychology
• ftable, ttable, ztable
• Students often forget their textbook
• Simanova
• Unequal Ns do not matter much when variances are
  similar
• Unequal variances do not matter much when Ns are
  similar
• When both are unequal the nominal alpha is
  meaningless.

20
21
Classroom Management

• We have a server that any student can access from
  anyplace with internet connectivity
• We have a cart with 25 very cheap laptops that
  basically work as dumb terminals to connect to
  server
• Many students bring their own laptops
• Biggest problem is to keep them from email/web
  browsing/game playing
• Apply cell phone policy to laptops
• Walk around room or have lab assistant sit at
  back to monitor

21
22
Classroom Management

• Lecture focuses on statistics lab on Stata--but
  some computer work during lecture
• Reinforces what they are being taught
• Keeps them awakeespecially if they are doing it
• Students can easily produce the correct
  statistics using Stata, but they have difficulty
  interpreting the results
• A statistically significant result can be trivial
• A confidence interval may highlight that a result
  is trivially different from zero effect
• Need to assess both substantive statistical
  significance
• Discuss pitfalls such as explaining rare events
• Simulations can help them understand assumptions

22
23
Room for Improvement

• corr does not give significance, just pwcorr or
  corrtab. Beginners do not understand why.
• Effect size post estimation command for all
  appropriate testsAPA expects to see effect size
  reportedwould open up a huge market
• Add icon on menus that will copy command at the
  end of the do-file, if only one do-file is open
• Enhance do-file editor
• Add professional quality three-dimensional
  graphics

23
24
Room for Improvement

• Missing values package including complex survey,
  longitudinal, multilevel data
• Develop factor analysis commands
• Nice to have some way that Stata could validate
  the most powerful user written commands
• Need book who how to systematically manage data
  using Statamore than just the commandsincluding
  the philosophy behind it

24
25
Room for Improvement

• Need menu icon in Stata to copy directly to a
  do-file
• Experienced users often have several concurrent
  do-files Stata wouldnt know which do-file
  should get the command
• Beginning students usually have a single do-file
• Single click option would be nice, when you have
  only one do-file open

25
26
Discussion Topics

• Experience gaining support for Stata
• Barriers
• Solutions to overcome barriers
• Ideas for Labs
• Favorite user written commands for beginners
• Instruction
• Data Management
• Analysis
• Suggested Stata enhancements to make it a better
  tool for instruction

26