Statistical Methods Crash Course Wanted
January 26, 2010 6:24 PM Subscribe
I need a quick course in statistical methods - Chi squared, Student's t-test, Kruskal-Wallis, etc. I have a research project coming up this Spring and I am sorely lacking in these specifics. Any book or website suggestions appreciated. More about the project inside...
I will be analyzing survey data about physician self-assessment of competency (you know, 1 for not competent, 5 for very competent, type responses) using Stata. We'll be looking for any association with other variables in the survey. Here's the paragraph from the research proposal that has me worried:
Data will be analyzed using both multi-level and longitudinal modeling techniques, as well as factor analysis, Chi-squared test, Student’s t-test, Kruskal-Wallis rank test, univariate correlation coefficients, and logistic regression, as appropriate. All variables that are considered important by the research team will be included in the regression model. Interaction will be formally tested by multiplying the two factors of interest (Factor A * Factor B) and then including this interaction term in logistic regression models. All statistical analyses will be reported using 95% confidence intervals and p-values.
I'm in medical school but don't have any formal statistical courses in my background. Any suggestions for a crash course in statistical methods?
I will be analyzing survey data about physician self-assessment of competency (you know, 1 for not competent, 5 for very competent, type responses) using Stata. We'll be looking for any association with other variables in the survey. Here's the paragraph from the research proposal that has me worried:
Data will be analyzed using both multi-level and longitudinal modeling techniques, as well as factor analysis, Chi-squared test, Student’s t-test, Kruskal-Wallis rank test, univariate correlation coefficients, and logistic regression, as appropriate. All variables that are considered important by the research team will be included in the regression model. Interaction will be formally tested by multiplying the two factors of interest (Factor A * Factor B) and then including this interaction term in logistic regression models. All statistical analyses will be reported using 95% confidence intervals and p-values.
I'm in medical school but don't have any formal statistical courses in my background. Any suggestions for a crash course in statistical methods?
You might want to check out the paper behind this and look at what references they're citing.
posted by Kid Charlemagne at 6:35 PM on January 26, 2010
posted by Kid Charlemagne at 6:35 PM on January 26, 2010
Best answer: I just had a workshop about stats in a research-development course. There is a website that will help you. Go down to Resources and try some of those websites. In particular, it looks like this online statistics textbook fits the bill.
posted by PercussivePaul at 6:40 PM on January 26, 2010 [1 favorite]
posted by PercussivePaul at 6:40 PM on January 26, 2010 [1 favorite]
Best answer: I love, love, love Gravetter and Wallnau's Statistics for the Behavioral Sciences. (Buy an old edition --- I got mine for $5 + shipping). It is eminently readable, and you can totally teach yourself with it.
Chrisamiller is exactly right. Use G&W to get up to speed (just work through the exercises chapter by chapter) and then use software to get the work done.
Seriously, I must have tried learning from six or seven texts before I was introduced to G&W and then things just magically clicked and stats became fun and understandable.
posted by fake at 6:41 PM on January 26, 2010 [1 favorite]
Chrisamiller is exactly right. Use G&W to get up to speed (just work through the exercises chapter by chapter) and then use software to get the work done.
Seriously, I must have tried learning from six or seven texts before I was introduced to G&W and then things just magically clicked and stats became fun and understandable.
posted by fake at 6:41 PM on January 26, 2010 [1 favorite]
Best answer: I am agreeing with fake. You really need to work through a statistics book to know what you are doing. Statistics is how we talk in the sciences and you need to learn the language if you want to be a scientist or interpret scientific results on behalf of your patients. Just running a test that someone else ran in their paper is like prescribing a drug to someone because it worked for your cousin.
posted by shothotbot at 6:50 PM on January 26, 2010
posted by shothotbot at 6:50 PM on January 26, 2010
A good statistics textbook is great, but in a pinch, I have also found Wikipedia to be a good source as well. Also, check out the sources in Wikipedia they can be really really helpful.
There are some good free statistics textbooks on the internet as well.
posted by lizarrd at 8:15 PM on January 26, 2010
There are some good free statistics textbooks on the internet as well.
posted by lizarrd at 8:15 PM on January 26, 2010
I second what shothotbot said. It's trivial to use a statistics program, but takes years of work to understand how to properly analyze data. If you want to analyze the data properly, find a graduate student or professor to help you with the project. This is not something that a few hours on a web-site can teach you how to do.
posted by eisenkr at 8:16 PM on January 26, 2010
posted by eisenkr at 8:16 PM on January 26, 2010
Best answer: For those who have said it is trivial to use a statistics program have not used STATA, which isn't some word wussy SPSS jobby but the real deal for doing serious work. At least that was what I was told when I was up on the mountain. (Mostly kidding: it IS different).
For STATA book: Statistics with Stata by Lawrence C. Hamilton.
STATA website: This fine bunch of info at UCLA is going to be your best friend
As for basics? How basic? Cartoon guide to stats is often recommended but it's on my self collecting dust.
Perhaps: Richard Harris's "A Primer of Multivariate Statistics" which isn't shy about it being, you know, MATH.
Most of the hand-holding books I am familiar with are for the social sciences, where a dread fear of calc has prompted many of them to avoid anything math related, and are not likely best suited for your purposes.
Better suited is campus resources - many schools provide support for stats so ask around - were I am at there are several different units even - (ie: Education has a gang focusing on item response and the Rosch model help, Math dept provides basic info etc etc)
posted by zenon at 10:32 PM on January 26, 2010
For STATA book: Statistics with Stata by Lawrence C. Hamilton.
STATA website: This fine bunch of info at UCLA is going to be your best friend
As for basics? How basic? Cartoon guide to stats is often recommended but it's on my self collecting dust.
Perhaps: Richard Harris's "A Primer of Multivariate Statistics" which isn't shy about it being, you know, MATH.
Most of the hand-holding books I am familiar with are for the social sciences, where a dread fear of calc has prompted many of them to avoid anything math related, and are not likely best suited for your purposes.
Better suited is campus resources - many schools provide support for stats so ask around - were I am at there are several different units even - (ie: Education has a gang focusing on item response and the Rosch model help, Math dept provides basic info etc etc)
posted by zenon at 10:32 PM on January 26, 2010
How about CMU's Open Learning Initiative, or MIT's OpenCourseWare?
posted by Mike1024 at 12:52 AM on January 27, 2010
posted by Mike1024 at 12:52 AM on January 27, 2010
Statistis for people who (think they) hate Statistics (Excel Edition) is a pretty simple presentation of stats.
posted by MesoFilter at 1:03 AM on January 27, 2010
posted by MesoFilter at 1:03 AM on January 27, 2010
At my institution, there is a biostatistics consultancy lab; the stat department also offers a consultancy service. As others have said above, the mechanics are not hard, knowing what is valid and meaningful is. You should see if your PI (I assume that there's a senior researcher somewhere in the project) has a relationship with stat people. You would really rather do it right the first time than have the stat reviewer at a journal point out that you did something wrong. Longitudinal modeling, multilevel modeling, and factor analysis aren't trivial. That said, the wiki pages and the stata manual are good places to start. By stata manual I don't mean the help topics or online pages; there is a huge bookset which accompanies stata with short articles explaining what's going on in either of their procedures.
posted by a robot made out of meat at 4:43 AM on January 27, 2010
posted by a robot made out of meat at 4:43 AM on January 27, 2010
Best answer: In the absence of a text book, I would read all of Doug Altman's Statistics Notes series from the BMJ.
My favourite stats book is based around SPSS not STATA, but still worth a look: Andy Field's Discovering Statistics.
posted by roofus at 1:37 PM on January 27, 2010 [1 favorite]
My favourite stats book is based around SPSS not STATA, but still worth a look: Andy Field's Discovering Statistics.
posted by roofus at 1:37 PM on January 27, 2010 [1 favorite]
Best answer: First, two book suggestions:
Intuitive Biostatistics by Harvey Motulsky (you aren't doing biostats, but it's great).
The Tao of Statistics: A Path to Understanding by Dana Keller. (It's conceptual, not mathematical).
And then one other suggestion, adapted from what my biostats professor used as the correct answer to a surprising number of statistical problems: Hire a statistician. (No seriously, add someone with stats chops to your research team. It will lend credibility to your work and give your team a needed additional perspective.)
posted by j-dawg at 2:03 PM on January 27, 2010
Intuitive Biostatistics by Harvey Motulsky (you aren't doing biostats, but it's great).
The Tao of Statistics: A Path to Understanding by Dana Keller. (It's conceptual, not mathematical).
And then one other suggestion, adapted from what my biostats professor used as the correct answer to a surprising number of statistical problems: Hire a statistician. (No seriously, add someone with stats chops to your research team. It will lend credibility to your work and give your team a needed additional perspective.)
posted by j-dawg at 2:03 PM on January 27, 2010
Response by poster: Thanks everyone. This are just the kinds of suggestions I was looking for. These are good resources. I am working with an experienced PI. I'm not trying to be the stats expert, I just need to be intelligently informed and able to work with the team.
posted by Grundlebug at 4:02 PM on January 27, 2010
posted by Grundlebug at 4:02 PM on January 27, 2010
« Older Experienced Sydney renter needs a little more help... | Affordable resources for rectangular tumbled glass... Newer »
This thread is closed to new comments.
posted by chrisamiller at 6:27 PM on January 26, 2010