Measurement Scales: Nominal, Ordinal, Interval

• Measurement refers to the assignment of numbers to characteristics of objects, people, stores, or events according to rules.
• There exist four levels of measurement scales: nominal, ordinal, interval, and ratio.

Nominal Scale

• Used in multiple-choice questions where choices have no sequential relationship and are mutually exclusive.
• Examples include gender (Male, Female) and province (ON, MB, QB).

Ordinal Scale

• Used to order or rank choices.
• An instance of this is ranking provinces from 1 to 5.
• The numeric values assigned only indicate rank/order, and the range between the numbers doesn't reflect the intensity of preference.

Interval Scale

• Scales with the same characteristics as ordinal scales, but the interval between the different choices is equal.
• This can be seen on a scale of 1-7 to rate the likelihood of purchasing a product again.
• Scales here have properties of order and difference, and interval scales have an arbitrary zero where 0 doesn't equal nothing.

Ratio Scale

• Has all the characteristics of nominal, ordinal, and interval scales, plus having a meaningful zero point.
• Examples include length, weight, age, income, and market shares.

Scaling Techniques

• Involve comparative and non-comparative scales.
• Comparative scales comprise paired comparison, rank order, constant sum, and Q-sort and other procedures.
• Non-comparative scales involve continuous rating scales and itemized rating scales.
• Itemized scales can further be comprised of Likert, semantic differential, and Stapel scales.

Comparative Scales

• Respondents are asked to compare items and rank each.

Paired Comparison

• Respondents are presented with two objects and asked to select one according to some criteria.
• Frequently used when the stimulus objects are physical products.
• It is easier to select one item from a set of two rather than a large dataset.
• This scaling technique has no order bias and the number of objects to be evaluated should remain small to prevent interviewee fatigue.

Rank Order

• Respondents are presented with several objects simultaneously and asked to order them according to some criterion.
• Frequently used to measure preferences for brands and attributes.
• More closely resembles a real shopping environment.
• It only produces ordinal level data.
• Results can be misleading if all alternatives in a respondent's choice set are not included.
• Being ranked may be outside a person's choice set.
• The researcher doesn't know why the respondent ranked the items as they did.

Constant Sum

• Respondents allocate a constant sum of units (points, dollars, chips) among a set of stimulus objects with respect to some criterion.
• If an attribute is important, it receives more points.
• Respondents may have difficulty allocating points to total 100 if there are a lot of characteristics/items.

Non-Comparative Scales

• Respondents make a judgment without reference to another object.

Likert Scale

• Respondents express their level of agreement or disagreement with statements by selecting from options like Agree, Neither Agree/Disagree, or Disagree.
• The technique uses multiple statements.
• Correlation analysis, reverse valence of statements may bias response.

Semantic Differential Scale

• A rating scale is anchored at each end by bipolar (Sweet-Sour) or monopolar (Sweet-Not Sweet) adjectives.
• 7-point scales are usually used, but 5-9 points can also be used.
• Uses multiple statements.
• A quick and efficient way of examining the strengths of a product versus competitors and is statistically robust.
• Lacks standardization and may have a halo effect.

Other Non-Comparative Scales

• Consist of graphic scales that mark opinion by marking a continuous line, from which numerical data is treated as interval data, but is not as reliable as itemized data.
• Disadvantages include extreme anchors that force respondents towards the middle of the scale.
• There also exist itemized scales, that select the best option from a limited number, easy to administer but offer no fine distinctions.

Balanced and Unbalanced Scales

• Balanced scales have the same number of positive and negative categories.
• Unbalanced scales are weighted toward one end or the other.
• If a researcher expects a wide range of opinions, a balanced scale is implemented.
• The degree of positivity given research findings is ascertained using an unbalanced scale.

Key Steps In Sampling

• Define the population.
• Identify the sampling frame.
• Select the sampling procedure.
• Determine the sample size.

Define the Population

• Specifying the characteristics of individuals for whom information is needed to meet research objectives.
• Consider geographic area, demographics, usage characteristics, and awareness measures.
• Define the characteristics of individuals who should be excluded.
• Develop screening questions.

Identifying the Sampling Frame

• Consists of the population/universe, and the sampling frame.

Population/Universe

• The entire body of units of interest to the decision maker.

Sampling Frame

• A listing of population units from which a sample is chosen.
• A sampling frame error occurs when the research population is inaccurate or incomplete.

Sampling Frame Error

• Targets that shouldn't be included are included and vice versa.
• Can be addressed using smaller samples and exercise quality control in interviews.
• A sampling frame is a list supplied by a commercial vendor.

Selecting the Sampling Procedure

• Divided into probability and non-probability samples.

Probability Samples

• All members of the sampling frame have a known probability of being included in the sample.
• Information is obtained from a representative cross-section of the population of interest.
• Sampling error can be computed and survey results can be projected to the total population.
• More expensive than non-probability sampling.

Non-Probability Samples

• The probability of inclusion in the sample is unknown for different members of the sampling frame.
• Researchers don't know the degree of representativeness.
• Sampling error can't be computed and results can't be projected to the total population.

Selecting The Appropriate Sampling Procedure

• Non-probability sampling includes convenience, judgmental, snowball, and quota samples.

Convenience Sample

• Uses people who are easily accessible.
• An instance is Frito-Lay using their employees.

Judgmental Sample

• The selection criteria is based on the researcher's judgment about what constitutes a representative sample.
• A common example of is product tests in shopping malls.

Quota Sample

• Quotas based on demographic factors are established for population subgroups.

Snowball Sample

• Additional responses are selected based on referrals from initial respondents.

Determining The Sample Size

• Sample size depends on client/researcher choices.
• Relies on acceptable margin of error, confidence level desired and questionnaire design-scales.

Acceptable Margin of Error Considerations

• Is a managerial decision for which a reasonable E needs to be picked to reflect actionability of greater when E is low and market research costs.
• Both are important.

Acceptable Margin of Error Formula

• For sample mean: n = z^2 * s^2 / E^2
  • E is the acceptable margin of error.
  • z is the standard normal variate value.
  • s is the sample standard deviation.
• For sample proportion: n = z^2 * p(1-p) / E^2
  • p is the sample proportion.

Estimating Methods

-Sample Standard Deviation (s):. -1. Results from previous search -2. Results from a pilot survey -3. Secondary data -4. Use judgement -5. Six sigma rule

Estimating Methods

-Sample Proportion (p):. -1. Same as 1-4 above -2. p=0.5 as a conservative estimate.

Sampling Definition

• Define the sampling frame to minimize bias (sampling frame error) and based on survey methodology.
• The sampling procedure should be conducted based on research task, and the sampling frame.
• The sample size should be calculated for each aspect.

Qualitative Considerations

• Importance of the decision, the nature and number of variables, nature of analysis like means or proportions and advanced analysis methods must be considered.
• Sample sizes used in similar studies, incidence and completion rates and resource constraints must all be accounted for.

Ways to Reduce Error and Increase Accuracy

• Reducing sampling error is expensive.
• Minimizing non-sampling error involves selecting an appropriate sampling frame.
• Incentives can be used to improve response rates and minimize non-response error.
• Design surveys that contain questions that are easy for respondents to understand and responses that are easy to record and code.
• An inverse relationship may exist between sampling and non-sampling error.

Ethical Considerations of PMR

• Concerns for ethics benefits PMR practitioners in three major ways.
  • Improves the acceptance of professions among and public, which directly impacts response rates and studies.
  • Prevents negative publicity and you reg
  • Allows researchers and scientists to keep their jobs.

Issues concerning respondent rights:

-No invasion of privacy - Obs studies in semi-public/private places - Qualifufive research techniques - Merging of dafa - Overly personal questions/topics

Ethical Issues

-Concern for subjects/respondents:.

   - Inconvinient/insensitive confacts
  - A neg physiological/psychological effect on respondents.
   - Nondisclosure

Research Ethics based on the TCPS-Tri Council Policy Statement

1. Research cannot involve more than minimal risk
2. Participants will be drawn from the general adult population, capable of giving free and informed consent
3. Research can not involve any personal, sensitive or incriminating topics/questions
4. Research cannot change or involve participants beyond the range of "normal" daily life
5. Research can not involve physically invasive contact with participants
6. Research cannot involve deception.

Recruiting Participants

• Potential participants must be informed of:
  • The purpose of the project and the data being gathered
  • The procedures involved in the research
  • The voluntariness of their involvement
  • Assurance of no harm, risk, or discomfort of their involvement
  • The confidentiality of their responses
  • The ability to withdraw from the study

Consent

• TCPS requires obtaining written, verbal or implied consent.
• Consent must be fully informed and free, with everything explained prior to the start of the study, with the absence of any coersion or undue influence.
• All participants must agree by signing a consent form or completing the survey after receiving and reading a letter of information.

Data Analysis Template

Single Variable Analysis of Sample Demographics

• Goal is to describe sample demographics and defend if the sample's representative
• Use frequency tables, graphs, pie charts, statistics (mean, standard deviation)

Single Variable Analysis of Remaining Questions

• Goal is to obtain a summary of responses to the questions and calculate the actual margin of error -Use frequency tables, graphs/pie charts, statistics (mean, standard deviation)

Multiple Variable Analysis

• Goal is to look for more intricate patterns of relationships between variables (cross fabs) -1. Select variables and develop hypotheses -2. Create a cross fable of frequency data -3. Calc expected values using Null Hypothesis -4. Sample size which must have at least 5 expected counts.

Inferring Causality

• In week 9, causality means a change in one variable will produce a change in another variable.
• Three types of evidence to infer causality are concomitant variation/association, time order of occurrence of variables and spurious association.

Nature of Experimentation

• Experimentation is the manipulation of one or more variables (cause) by the experimenter in such a way that its effect on one or more variables can be measured. -Manipulation -Measurement.

Experimentation Variables

• Independent variable (variables manipulated by experimenter)
• Dependent variable (var that will reflect the impact of the manipulations) -Treatment Group: Portion of sample exposed to the manip -Control Group: Portion of sample for whom independent war is unchanged

Notation

• X: Exposure of a sample to the independent var
• O: Observation of measurement of the dependent var -Movement through time is represented by a horizontal arrangement of X's and O's from left to right -Simultaneous exposure/measurement is represented by a vertical arrangement

The Concept of Validity

• Internal validity: the ability of the experiment to unambiguously show a cause and effect relationship. -External validity: the extent to which the results of the experiment can be generalized to other people, settings, and time.

Threats to Internal Validity

• Is the change in the dep var due to the indep var or due to something else
• "these 'smfh elses' represent the threats to informal validity."

History

-Any var/events other than the ones manipulated by the experimenter that occur before the pre- and post-measures and affecting the valve of the dependent variable. -One group: is there a history effect? -Multiple groups: is there a differential impact of history for two groups?

Experimental Mortality

• One group: is there a loss of respondents within the group bofw the pre and post-measures?
• Multiple groups: is there a differential loss of respondents across two groups between the pre and poof measures?

Threats That Concern Study Quality

Maturation

• The biological/psychological process that systematically varies with the passage of time, independent of the experimental variable (older, hungrier, fired). -One group: is there a maturation effect between the pre and poof measures? -Multiple groups: is there a different maturation effect for two groups, bofw the pre and poof measures? -The longer the experiment, the more likely this may become a prob

Main Testing Effect/Premeasurement

-An effect that is a by-product of the research process itself. -One group: is there a main testing effect within the group? -Multiple groups: is there a differential main testing effect across two groups?

Instrument Variation

-Changes in the measuring instruments that might account for differences in measurement (changes in scales, changes in interviewer). -One group: Is there instrument variation within the group? -Multiple groups: Is there variation with the i'nst var with in and/or across groups?

Selection Bias

• When the groups formed for the exp are initially unequal with respect to the respondent variable or in the propensity to respond to the independent variable.

Remedies

-Randomization -Matching

Interaction Effect

• When a premeasure changes the respondent's sensitivity or responsiveness to the independent variable.

Threats to External Validity

-All previous discussion from int val t -Surrogate situation: when the env, pop samples and/or the treatment effects are different from the actual situations. -Measurement timing: when pre/post measurements are made of an inappropriate time to indicate the effect of the experimental treat mont.