PSYC2009 Study Guide - Quiz Guide: Statistical Inference, Sampling Frame, Random Assignment
Sampling
• Sampling questions
o What kind of sample?
o How to select sample to avoid bias?
o How big should the sample be?
o How close Is your estimate likely to be to the population percentage
o What degree of confidence should you have in this estimate?
• Population: the whole group
• Sample: the bunch you select out of the population
• Population perimeter:
• Need to make sure sample is representative of entire population: randomised sampling
• Randomised assignment of subjects to the different treatment conditions enables the
experimenter to use inferential statistics to ascertain whether the differences may be real or
happenstance
o Random groups differ, they won't have the same means or standard deviations
• The only other way would to be exactly match people (same age, same symptoms,
same gender, etc.) - this can work and is still used, but generally random
assignment is used more
o Randomized samples usually randomise differences between the groups
• Sampling procedures all have two elements:
o A sampling frame which defines the population from which the sample is taken, and
o A selection procedure by which the sample actually is taken.
• Desirable properties of samples for inference to population from which sample taken:
o Representativeness in the sampling frame
o No bias in the selection procedure.
• 2 kinds of samples: probabilistic and non-probabilistic
o Probabilistic if for inferential statistics
• Simple random and systematic
▪ With replacement
▪ Without replacement
▪ Each person in the sampling frame is ordered and selected at random, or
every "kth" person is selected
• Stratified random
▪ The population is broken down into different groups (strata). A given
percentage of each group or strata is sampled
▪ Reduces the sampling variation (sometimes even to 0 but not always)
▪ Usually the best way to go, as it reduces this random variability
• Cluster sample
▪ Multistage cluster: the population is broken down into different groups
(clusters). All units within some clusters are sampled
▪ Randomly sample clusters first and then narrow it
• Randomly select schools in Canberra
• Randomly select classrooms in each school
• Randomly select 15 people from each room
• 3 layer cluster sample
• Inflate the variability in the mean
• Problems with random sampling
o Requires complete enumeration (listing) of population sampling frame
o Refusal rates can seriously compromise/destroy result representability
o Sometimes may be ethically/politically unviable
• Nonprobabilistic is for descriptive or sometimes exploratory purposes.
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
This can work and is still used, but generally random assignment is used more: randomized samples usually randomise differences between the groups. Simple random and systematic: with replacement, without replacement, each person in the sampling frame is ordered and selected at random, or every kth person is selected. Stratified random: the population is broken down into different groups (strata). All units within some clusters are sampled: randomly sample clusters first and then narrow it, randomly select schools in canberra, randomly select classrooms in each school, randomly select 15 people from each room, 3 layer cluster sample. Subject to eligibility criteria: convenience, 1st year psych students are a good example - available and rewarded, haphazard (but not random!, does(cid:374)"t look syste(cid:373)ati(cid:272), (cid:271)ut it will (cid:271)e at least su(cid:271)(cid:272)o(cid:374)s(cid:272)iously i(cid:374)flue(cid:374)(cid:272)ed (cid:271)y your own preferences. Snowball: need to interview drug addicts for certain experiments. You may find 1 or 2 and then ask them if they know anyone else who is also addicted.
