Q9 ch28/29 data management+descriptive stats

data management

entire process of record keeping before, during, and after res. study

function of a codebook?

a guide written for a spef. study- describes each variable + how collected data will be entered into a computer file

also describes how anticipated data problems (missing responses) will be handled

spreadsheet

file that stores data in cells… row-by-column table

database

data man. sys. stores data in tables which each row represents one record represents one record - related records in diff. tables can be linked

double-entry

ensures accuracy - two individuals enter same data into separate computer files → compare → resolve discrepancies

data cleaning

correcting any typographical/other errors in data files

derived variable

new variable created during data analysis from existing variables in the data file

recoding

generating values for a new variable based on 1+ existing columns of data in a file

*always save backup version of data file before creating new variables in duplicate copy

how does stat software programs help w res. process?

1) most software programs can run all common stat functions

2) results from diff programs are identical/nearly

general data security process for computer files. paper files?

paper records- safely store paper records- (+ signed informed consent statements in locked n secure room)

all files w sensitive info should be password protected + limited access only to essential research personnel

descriptive stats

means. medians, proportions, standard deviations, used to characterize distributions of quant. data

general analytic plans: case series

univariate analysis- describes the study pop

general analytic plans: cross-sectional

unvariate + a little of bivariate

describe study pop + compare groups

general analytic plans: case control/cohort/experimental

univariate+bivariate+ a little of multivariate

describe study pop + compare + regression/other adv. analysis

univariate analysis

one variable in data set- uses simple stats (counts aka frequencies, proportion, averages)

bivariable analysis

rate ratios, odd ratios, other comparative stat tests used to examine associations between 2 variables

multivariable analysis

stat tests such as multiple regression models - used to examine relationships w 3+ variables

variable

characteristic assigned to 1+ value

ratio variable

numeric variable, able to be plotted on a scale

value of zero indicates total a sense of characteristic

ex: income, height, weight, temperature

interval variable

numeric variable + has order that is meaningful

zero does not indicate total absence of characteristic

ex: IQ score, temp in F or C, college years

continuous variable

numeric variable that can take on any value within a range

*can be added on any point on a plot

discrete variable

numeric variable can ONLY take point values, no value in between

ex: ppl in this class, eye color

ordinal variable

aka ranked variable, responses that have order but are not equal/systematically meaningful

ex: tier rankings, best to worst, first to last

nominal/categorical variable

has values but has no inherent rank/order aka a list of stuff

dichotomous variable

only two possible answers

binomial variable

dichotomous variable- coded as having values of only 0 + 1

categories are mutually exclusive

central tendency

types of average values- mean, median, mode

mean

add up all values, divide the sum by total number of individuals w/ value for variable

median

put all values from least → greatest, find middle number

mode

most frequently occurring value in a data set

how to describe central tendency for ratio + interval? ordinal? categorical?

ratio + interval - means, medians, mode

ordinal - median/mode

categorical - mode

histogram

graphical representation of distribution of ratio + interval data

x-axis - values of responses

y-axis - count of # of times each response appears in data set

bar has to be same width, no gaps between bars

boxplot

aka box-and-whisker plot, graphical depiction of numeric variable

displays median, interquartile range, outliers

bar chart

graph presents categorical data using equal width rectangles, can appear in any order

pie chart

circle, each wedge/slice shows percentage of participants w/ a spef value

normal distribution

aka gaussian distribution- bell shaped curve

kurtosis

how peaked/flat a distribution is

leptokurtic - looks like a peen, super big in the center

platykurtic - small hill in the middle

skewness

asymmetrical of normal dist.

variance

add squares of differences between each observation + sample mean → divide by total number of observations

standard deviation

square root of variance

z-score

how far from the mean each data values are, using standardized scale

score - mean/ stand. dev.

standard deviations under normal distributions: 68%

68% of responses fall within 1 SD above/below mean

standard deviations under normal distributions: 95%

95% of responses are within 2 SDs above/below mean

standard deviations under normal distributions: more than 99%

more than 99% of responses are within 3+ SDs above/below mean

small SD? large SD?

small SD- data points are tightly clustered around mean/ data has similar value

large SD- data points are more scattered + more varied

descriptive stats

describes basic characteristics of quant. data (means+proportions)

typical reporting: ratio+interval w/ normal dist

mean + SD

typical reporting: ordinal ratio+interval w/ non normal dist.

median / interquartile range

typical reporting: categorical

proportions of participants who provided particular responses

confidence intervals provide info ab

expected value of a measure in a source pop based on value of that measure in a study pop

sample size relates to confidence interval

larger sample size = narrower confidence interval (more room for errors)

95% confidence interval

5% of the time, CI is expected to miss capturing true value of a measure in source pop

three most serious forms of research misconduct

fabrication: creation of fake data,you made shhi uppp

falsification: misrepresentation of results

plagiarism: use of other ppls things

main benefits of consulting w a statistician early in st design process

ensures that:

sample methods/size are appropriate

questionnaire will yield usable data

analysis plan is reasonable