1. ~Survey
- set goal - what you want to learn (why, what or how); watch out that correlation does not imply casuality (reason-cause relation);
- sample - determine who will be interviewed; the bigger sample the better; try to avoid bias (nice people, customer people, older people, ..);
- methodology - how you will interview: personally, by telephone, by mail, by email, as online survey (each has different risks of bias or other disadvantages);
- questionarre - create it; useful rules:
| ----> | * start with introduction or welcome message, then explain the reason for the survey (shortly); why it is worth taking it, how it will influence our lives, etc. * reassure confidentiality * instructions at the point that needed, not earlier * early questions should be easy and not too personal * avoid technical terms * change question types not to make people not think about what they are doing * keep it short * PRETEST! QUESTION TYPES: - open-ended - closed-ended * include "none" and "other" * include most possible choices to avoid bias - Likert scale * start from "Strongly agree" - rating scales * 1..10 or 1..5; or 4..10 - in Finland * higer numbers more positive * include "non applicable" |
- collect and analyse results
- communicate results
2. Controlled experiment
- in general: construct some cause-effect relation, then apply some treatment and make observations;
HYPOTHESIS
Every study must have at least 2 hypotheses (because they must cover all possibilities):
H0 - the treatment has no predicted effect
H1 - the treatment has the predicted effec
Types of hypotheses:
- One-tailed - specifies direction of the difference
- Two-tailed - does not specify direction of the difference
We should have expetimental group of one or several levels of treatment plus control group against which we will compare the treatments, the treatment there can also be "no treatment".
- Between subjects design - each subject exposed to one treatment level; can be done with pre-test - measure the dependent variable also before the treatment;
- Within subjects design - each subject exposed to more than 1 level of experimental treatment; but has some drawbacks, eg fatigue effect, learning effect;
STATISTICAL METHODS
Types of variables:
- nominal, eg. male/female, Java/Python/C;
- ordinal, eg. Likert scale (but intervals may be uneven);
- interval, eg.Celcius temperature;
- ratio, eg. Kelvin temperature (there's logical 0);
- central tendency - mean, median (middle value), mode (most frequently occuring)
- dispersion - range (=max-min), variance, stardard deviation (sqrt(variance));
- distribution - frequency, percentiles, charts, plots
** TODO
3. Mathematical modelling and simulation
4. Theorem proving
5. ~Participant observations
6. ~Grounded theory
