Saturday, February 21, 2009

Surveys and Sampling

Surveys are a type of descriptive research that allow researchers to focus on a few variables within a sample, a smaller yet representative group, of the larger population. In this way, surveys enable researchers to collect decriptive information about readily observed or recalled behaviour within a large group or population of compositions, courses, teachers, or classrooms in terms of the sample.

In surveys, the subjects under investigation are referred to as a sample. In order to determine which subjects will be part of the sample one must:
1) Determine large population from which sample's to be drawn
2) Determine size and type
*takes into consideration (confidence limits / precision- what degree of
imprecision is acceptable?)
Once the previous 2 points have been addressed one can go on with the actual selection of the individual subjects.

Random sampling : easiest - often referred to as the best way
* number the population (adhering to rules for listing)
* consult the chart of random # to draw sample - which involves - point at which
researcher will start table, direction of reading
**Charts can also be made by calculator or computer with random number function**

Systematic Random Sampling
* Must have ordered/organized subjects (interval unit) in place.
* Observations to start at point picked at random in the ordered unit
* Must examine population for any natural or planned order in it
* Complete population must be accounted for. size of population/size of sample
unit = desired sample size.

Quota Sampling
* A population and known characteristics are identified and then samples picked
that have corresponding characteristics
* Helpful when one knows percentage of incidence of specific features (such as
sex / race/ etc) of a population

Stratified Samples
* Used when some part of population are of more interest than others.
* Main population stratified - researcher focuses on strata of most interest
* Allows for a closer look at specific characteristics within a population
* Potential downfall - Due to size of strate in comparison to population and the
other strata, confidence limits may exceed that which is acceptable for

Cluster Samples
* For use when one wants to study individual units within a large populaiton but
doesn't want random b/c of potential difficulties with other variables.
* AVOID unless you have a statician to aid as a consultant.

Data can be collected using almost any type of data collection methods. It can be collected in the form of questionnaires, paper collection, interviews, test results, etc. (63) In the case of questionnaires and surveys there are two type of questions one might use: open ended or objective scored questions (aka multiple choice)

objective scored questions are succint, easy to analyze, standardize, and comparable whereas open ended questions are longer in length, allow more variance in the response, are less predetermined, more difficult to analyze, and can be limited by writing ability and time of person filling it out.

If one chooses to use questionaires or surveys the following must be adhered to:
Editting and revising questions for directness, simplicity, clarity by way of
* Submitting them for review by others
* Using a small pilot sample of population for review of intial responses

It is important to kp in mind the common problems of open ended questions - ambiguity, asks information that respondent doesn't have/know, poses suspicious questions.

Once the information is collected analysis can ensue.
One must determine the number of variables represented as a "K" and the number of
sample units (aka number of subjects within sample) represented as n.
Their relationship is plotted using a rectangular matrix.

It should be noted that "n" will by much larger than "k" b/c one wants to study a few features of a large group that's been reduced to a sample.

The analysis of information collected from a sample depends on the type of information collected.
Three types of data that can be collected are: nominal, interval, and rank order.
Nominal - simple counting and/or percentages (easiest to table)
*Type of nominal is frequency data - percentage and proportion
(more convenient for sampling studies)
Interval: come from test scores w/ large number of items, ratings, grades (can calculate confindence limits or precision of results (More difficult to table)
Rank order - type of interval - subjects/compositions/teachers placed in heirarchal order and assigned ranks from 1 - n (equal intervals are ASSUMED)

One can analyze data collected in terms of range / standard deviation/ variance
Range : (highest score of a variable - lowest score of a variable)
Standard deviation = to 1/5 ti 1/6 of the range
Variance : measure of differences b/n scores - square of standard deviation

for nominal data expressed in percentages - standard deviation is derived from mean
for interval data - precision related to standard deviation on variables of population.

Generalizations can only be made from sample to the group that the sample represents. In doing so, the generalization must account for the confidence limits of the variable(s) under investigation.

1 comment:

  1. Hi Nik,
    Good post, by the way. You gave me a lot of good information here. In particular, you have a good focus on constrains imposed by the group represented by the sample. There seems to be slight of hand occasionally. In other words, the researcher samples one population (let's say Dutch students at one level at one school) than attempts to broaden the results for a larger population (all students). It just doesn't hold water.

    We're in agreement, it would appear.