Statistical Analysis
Basic Statistics
SurfStat
Australia (Hypertext stat book--very good)
Stat
Notes--excellent summaries of stat techniques.
Research
Methodology Links.
Research
Methods Knowledge Base-Bill Trochim's excellent online
text
WWW
Resources for Statisticians.
Survey
System (Exceptional online calculator for sample size, confidence
level, and confidence interval)
InfoPoll
Library (templates for surveys and polls)
The Poisson Pit
(online poisson distribution calculator)
Random
Number Generator (creates lists of random numbers within a range)
Calculators
Online (over 6,800 calculators for all occasions in most fields)
Trochim's
Stat Page (links to calculators, tutorials, tables, etc.)
Research Methods
& Statistics Links (Wesleyan collection)
Social Research
and Statistics Links (USI Dept. of Sociology links for qualitative
research, scaling, ethics, grant writing, sampling, etc.)
Which Statistics To Use
STATCON:
Artificial intelligence statistical consultant prompts you for info about
your data and then recommends the proper statistic.
How
numbers are tricking you. Use and misuse of stats.
Data Entry & Processing
Basic Stats
Your selection of an appropriate statistic should be based on the characteristics
of your data (e.g., sample size, distribution, etc.) and what it is you
want to test. The following statistics are most often used on theses/FAP's
and you might consider them:
-
Descriptive Statistics are the most basic and involve using numbers
to characterize your sample. Typically they use means (or medians), standard
deviations (or other variances), and percentages. Sometimes, these are
cross-tabulated by some other variable such as age, gender, or worker characteristics.
-
Correlations measure the degree of relationship (not causation!)
between two or more variables. That is, if one goes up, the other goes
up (direct relationship), like promotion and salary usually covary. In
contrast, some variables have an inverse relationship in which one goes
up and the other goes down. These values are expressed from -1.00 (inverse
relationship) to +1.00 (direct relationship). When one variable is used
to predict another, a correlation can be expressed in a regression equation;
that is, a score on one measure can be used to predict the score on another.
For example, the SAT and GRE are used to predict academic performance.
Be careful with very large numbers of participants (>500) because you can
get significantly high correlations that are more a function of sample
size rather than the variable or experimental condition itself.
-
T-test is a measure of the difference between means of the same
or matches samples. It uses the mean and standard deviation, and therefore
the sample should appear normally distributed (bell-shaped curve). If you
test more than three means, you increase the chance of finding an accidentially
("spuriously") significant t value, and therefore should use an ANOVA (analysis
of variance) for more comparisons.
Deciding Which Statistic To Use
If you are doing a quantitative study, your advisor will help you
decide which statistics are most appropriate to use to describe your data
and test your hypotheses. In discussing these issues, you should be familiar
with the following information:
-
What is the level of measurement of the variables with which you
are working (e.g., nominal, ordinal, interval, or ratio)?
-
Which variables are dependent and independent? Are there any intervening
variables that might contaminate your data, or confuse exactly what
you're measuring?
-
Which variables or combinations of variables are you wanting to compare
(that are directly related to your hypotheses)?
If you are interested in exploring different statistical tests and want
to use a self directed computer program to help you, a wonderful little
program called STATCON (statistical consultant) is available
as shareware. It prompts you to answer simple questions about your data,
then recommends which stat to use! At CSS, Statcon is available from Dave
Swenson (bring a disk to download a shareware copy). An online version
of a similar program is available at:
Data Entry and Processing
If you want to process your own data and have an IBM compatible computer
at home you might want to purchase the student version of SPSS/PC (about
$45). Most basic functions can be tested using this home version. The larger
version of SPSS is available in the CSS Computer Lab for nearly any statistic
desired. Check with your advisor before you make a purchase decision.
Consider using computerized answer sheets for subjects to respond to
questionnaires. This data can be easily scanned onto a disk for statistical
analysis. Otherwise, someone will need to enter data by typing each score
into a database. DON'T enter your data into a database until discussing
this with your advisor, since some databases are not compatible with all
statistics programs. Research assistants may be available to help
you enter and process data ($20-30 per hour). Your advisor will meet with
you to help interpret the printouts.