Data is a Dish Best Served Raw

1
Data is a Dish Best Served Raw

• The Art of Feeding a Computer
• Sam Butler, MS
• Ralph OBrien, PhD
• CCF Biostatistics

2
Data Fit for Humans

• Your initial spreadsheet is set up for your
  convenience as it should be.
• When you give the data to a statistician you are
  asking that individual to use a machine to
  extract information from your data

3
Data Fit for a Human
4
Some Things about Machines

• Computers are color blind
• Computers are not literate
• Computers cannot think, reason, interpret, second
  guess, intuit, infer, extrapolate, read your
  mind, or know-what-you-meant-to-write

5
Feeding the Computer
6
Things to Never Do

• Do NOT include
• Comments computers cant read
• Summary statistics data analysis is performed
  on raw data
• Shading, Coloring, or Bolding
• Graphs computers are digital not analog
• Symbols !_at_()\ etc.
• Poor codes for missing data N/A, -999, ? etc.

7
More Things To Never Do (Remember HIPPA!)

• No Patient Names
• No Patient Social Security Numbers
• No Patient Addresses
• Patient Data is confidential and you must insure
  your data will not violate privacy rules.

8
Correct Data Entry

• Data must be in column format
• Each COLUMN lists a single different variable.
• One column one entry
• Therefore each ROW is a single record with values
  for several variables for a single subject.
• There may be more than one record for each
  subject (e.g. multiple visits), but if, so then
  each record must contain the patients ID value.

9
Data Entry - Dates

• Record dates in a consistent date format
• 01/06/04 is this Jan 6, 2004 or June 1, 2004
• Use obvious variable names
• Birthdate, TreatmentDate, FollowUpDate
• Let the computer use these dates to accurately
  compute elapsed times of interest
  AgeWhenTreated TreatmentDate -
  Birthdate

10
Data Entry-Coding

• Gender and yes/no questions
• Troublesome gender 1 female, 2 male
• Right gender f female, m male
• Often best female 1 , 0
• 1 then means female and 0 means not female
• OK CarriedPregnancy 1 yes, 0 no
• Better? Nulliparity 1, 0
• 1 never carried a pregnancy,0 has carried a
  pregnancy

11
Data Coding Preference

• Statisticians debate this point, but Ralph
  advises to code categorical data with obvious
  values
• Bloodtype A, B, O, AB
• Instead of
• Bloodtype 1 A, 2 B, 3 O, 4 AB
• Dose 00mg, 25mg, 50mg
• Instead of
• Dose 0 00mg, 125mg, 250mg

12
If Written Entries are Unavoidable

• Record written information consistently.
• Florida, Fla, Fl, Fla., Florida and 3 other
  places
• Watch out for case sensitivity.
• Race W,w,B,b, 4 different races?
• Watch out for zeros vs. ohs and ones vs. els
• Treatment t0, tO, t1, tl 4 different
  treatments?

13
Data Entry Additional Coding

• Leave missing or not applicable entries blank or
  use your softwares standard coding for
  missingness, e.g. SAS uses a single lone period
  (.).
• Separate code for non-response.
• Non-response is not the same as missing data.
• If it is treated as missing it may drastically
  alter your conclusions.
• SAS allows .a, .b, etc. for such coding, but it
  will treat it as missing by default.

14
Well Prepared Computer FoodExcel ltgt JMP

• Variable names in first row max 32 characters.
• Never start a variable name with a number.
• Avoid using special characters in variable names
• None of these - !_at_()?lt gt \ /
• For variable names with more than one word use _
  between each word or capital letters
• Example entry_date, EntryDate
• Make sure date fields are formatted as date
  fields, numeric fields are formatted as numeric
  fields, etc.

15
Saving Excel files

• If you have an Excel WORKBOOK with more than one
  worksheet you will have to save each sheet
  separately

16
Built in Excel
17
Opened directly in JMP
18
Whats in it for you?

• In far too many studies, data cleaning and
  manipulations that could have been avoided
  account for far more time than actually doing the
  statistical analyses.
• Build quality right into the design of your data
  files. This will allow everyone to quickly focus
  on analyses and interpretation.
• Doing it correctly at first will save time in the
  long run and help prevent costly mistakes, some
  of which may never be found Work smart and hard
  at first, or pay dearly later.