Sampling Errors

Difficulties of Kinds of Samplings

Non aleatory or non random

Probabilistic (in some degree)

Concern

Values

Opportunistic

Voluntaries

Choice

Quotas

Just random

Systematic

Cluster

Stratified

Definition

First arrived at central place

Selected by themselves and ethic

Focusing with criteria on special people

Random first entrants, up to complete a given representative quota of group or strata.

Organized randomness to equal probability of each to make part of investigation

Since the first one selected, followers are systematically pick each given number of subject.

Select group, then individual of each group. Group practically defined with no influence on investigation.

Select individual randomly in group reestablished to be more homogeneous (so influencing investigation)

Subjects

Individual

Unequal chance to be selected

Easiness or ethical criteria.

Uncertainties on the issue

Special characteristic

Weak occurrence

Intent to make randomness in selection under a previously weighted, partition

Equal probability to be selected. Reposition rules (even if twice registered).

The first one defined much. Parallel periodicity can bias all.

Compromise between a given phenomena and it’s following and easiness (

Compromise between a general investigation and group of concern

Group

Special according place of investigation

Inconvenient (or expected group effect)

Allowed by proceeding (place selection),

Goal of ideal subgroup.

Defined by a population of reference

Neutral. If group to check, have the proper categories and variables.

Check systemic criterion to be much smaller then number of group.

Practical interest, to limit the inconvenient for the whole population

Of special interest but inside population

Population

Confusing, find the lack of effect of sampling

Confusing and selective

Contradictory with the sampling

Supposed to better representativity.

Main goal, method of reference

Dubious. Control

Try to make easier estimates for the whole population.

Compromise group-population

Purpose

Pragmatic

Easiness find adjustment to

Easiness when ethical issue. For any investigation in all human one there is always this criterion.

Qualitative design or complementary, focused research (understanding the set of characteristics)

Best pragmatic compromise with investigation of population, if no other bias is introduced.

Randomness is not so easy; it requires neutralizing the observer effect.

When enough knowledge for to escape bias and insufficient for to choose a better procedure.

When quota method too suspicious, find a more neutral but economical way

When finding groups of intervention (knowledge and eligibility)

Mixed

No very proper

Same fundamental purpose, for to satisfy a fundamental need

Not very proper

Proper use

Reference

Perhaps for unspecified curiosity or indirect concern.

Good for half neutral assessment (in good conditions)

Good compromise

Fundamental

For to find qualitative undiscriminating effect

Preferred, experimental situation. More fundamental level to referred

Main interest, defined well a special group under a special phenomena

Study of mass phenomena, not fully trustable

Reference for mass or population phenomena.

Openness or undirected curiosity. Fundamental metrology and benigness of those.

Not the best, not very fundamental. General purpose

Not the best, good with high number

Conditions

Reproduction or longitudinal

Better repetition follows with quota method

Preferred goal (following effect). Demonstrated benefit or need.

Completed with general investigation of occurrence, (with very specific indicator).

Second best choice, especially for complex following, repetition of transversal

Neutral. Study distribution to better define significant number of sampling.

Just for easiness, or even placebo effect, when confirmed the lack of periodicity. (sort of military rule)

If the number if enough (large scale, general simple data).

Some appropriate procedures use it: sentry sites for to follow given phenomena

Transversal

First glance

Not very useful

Main interest, for to make definition

Most common choice, with enough effective (hundreds)

Matching good and proper tool with good sampling.

Have curiosity in other kind of methods (for to control)

General population data identification of structure

Operative choice of special thing (exposed) or program

Geography

Vice, easiness or best opportunity: find a central place to make a rapid assessment.

Generally special place, and inconvenient

(lost of view)

Can be the special reason or at the opposite non effectual.

Try to neutralize the effect of central place if disturbing representativity.

Precisely to neutralize thanks to randomization

Control systemic bias on the period of election & space structure

Generally the criteria of simplification. Can oblige to coarse categories.

Better if selections of group criteria are operative. Problem if general shared

Conditions of liability

Representativness

Poor, possible

Bias

Poor,

Possible bias

It is not the purpose, except for the ideal subgroup

Said to be the best in non randomized sampling

Most representative.

Check according criterion and complementary knowledge.

Have to be check, including absence of effect of criteria

Have to be check for scale or aggregative effect on population

Uses

Easiness. When lack of first estimates, to caliber following investigation. Qualitative detection. Categories design

Experimental authorized methods. For to qualify qualitative issue (fundamental binary value)

For to describe phenomena in ideal subgroup (make a fundamental reference).

Basic of population investigation. Problem with stability or non rational opinion or reaction to enquiry

Easiest calculus.

Methods of theoretical reference.

Simplicity

To make more easy huge investigation. Some optimum procedure to better representation

To better the information on given group, said at risk or targeted

Cautions

Select neutral place. Have enough parameters and numbers in each categories

Ethical control. Ethical reserves. Good for witness case. Blind investigation.

Good for witness case. Double blind investigation.

Considering marginalization.

Democratic-ethical concern.

Autoregressive problem

Randomization needs good procedure: basis, randomization, check non answer.

Qualitative complementary investigation. Detect if bias could have been introduce

Better with general characteristic investigations

Good structure of population, extend calculus of estimators to the whole population