WILLIAM HOLMES' METHODOLOGICAL TIP ARCHIVE  Last update: 10-10-2011
Scroll through or select a topic.
Charts 
Coding
Computer Files
Data Analysis
Data Entry
Ethics 
Findings
Hypotheses
Literature Review
Measurement
Problem Formulation
Sampling
Tables

CHARTS

• Three Dimensional charts can distort comparisons between groups if the apparent volume of the 3D element does not maintain the same proportional relationship to the other 3D elements as it would have as a 2D figure.
• A color chart is easier to read if bright colors are reserved for the categories or information you wish to emphasize or compare.
• Pie charts work best when you only have a few categories to compare and no more than one category represents a small percentage of the cases.
• A bar chart of a categorical or nominal level variable often looks better as a horizontal chart. A bar chart of an interval or ratio variable may look better as a vertical chart.

ToTOP

of the page
CODING VARIABLES

• Using integer codes usually gives more flexibility in manipulating and recoding variables than alphabetic characters.
• Use numeric codes instead of alphabetic codes. They are easier to work with unless alphabetic codes are needed to convey substantive information (e.g., a person's name).
• When coding ordered categories, it is better to assign higher numbers to the categories that have more of a characteristic or property and lower numbers to those that have less. It creates less confusion when analyzing data.
• When deciding on codes for missing data, draw a distinction between not applicable and not available or unknown. You may be able to substitute estimated values for unavailable data. You shouldn't do that for inapplicable data.

ToTOP of the page
COMPUTER FILES

• Always make a backup of your computer files. Keep a copy at a location apart from the original file. Do not keep your backup on the same computer as the original file.
• When naming computer files, give them substantively meaningful names that suggests the content and format. They will be easier to find and remember.
• Give all the files associated with a project a common name element that tells what project with which it is associated.
• A shorter file name is preferable to a longer file name.

ToTOP of the page
DATA ENTRY

• If the number of possible codes for your variables is limited, then valid range checks may catch most data entry errors. If the number of possible codes is large, then double-entry verification may be needed to catch most errors.
• Eyeball scanning of the data (also called Internal Visceral Sort) may identify format errors not otherwise noticed.

ToTOP of the page
DATA ANALYSIS

• You can tell whether two variables are related by either applying a statistical significance test with a probability sample or by applying a substantive criterion to a measure of association.
• Inferential statistics tell you about the presence of a relationship; measures of association, how strong the relationship is.
• Choosing a Measure of Variation If you want the most extreme difference, choose the range. If you want to know the difference for the preponderance of the cases, choose the interquartile range. If you want to know the average difference, choose the standard deviation.
• Choosing a Measure of Central Tendency If you want the average value, choose the mean. If you want the most typical value, choose the mode. If you want the most intermediate value, choose the median.

ToTOP of the page
DESIGN

• Experimental designs are best for demonstrating causal effects -- provided they are feasible, ethical, naturalistic, and implemented well.
• Quasi-experimental designs are best for demonstrating association between variables -- provided they deal with history, maturation, testing, and suppressing or reinforcing causal influences.
• Non-experimental designs are best for describing characteristics and raising questions -- provided the external and internal validity is justified.
• A mixture of designs can sometimes be cheaper and easier than using only one design for everything.

· ToTOP of the page ETHICS

• The more constraint on refusal to participate, the more limited should be the inducement offered.
• Don't use data from an unethical source (Don't eat fruit of a poisoned tree).
• The health and safety of research subjects has priority over validity and reliability of the data.
• Always check if the ethical standards for organizations involved in your research are more stringent than those of the Institutional Review Board approving your project.

ToTOP of the page
HYPOTHESES

• A well stated hypothesis can be reworded into a clear research question, and vice versa. If rewording your hypothesis as a question produces an ambiguous question, the hypothesis needs revision.

ToTOP of the page
LITERATURE REVIEW

• When multiple sources start having the same references and citations, it's probably okay to stop looking for new sources -- unless you're doing a dissertation.
• Summarise common themes from your literature together. Do not list what each source says separately, unless they all disagree and have no conclusions or findings in common.
• All quoted material authored by others must have a citation for the source that includes the page number of the quote (if any).

ToTOP of the page
MEASUREMENT

• To tell if something is a variable, ask yourself if you can name more than one characteristic, value, or attribute of it that share soomething in common. For example, rage and annoyance share anger in common; short and long share length in common.
• Four principles of good measures are that they should be: 1. unidimensional, 2. exhaustive of the characteristics measured, 3. possessing mutually exclusive categories, and 4. have clear definitions.
• When writing questions for a questionnaire, have each item refer to only one single thing per question. Otherwise, the response may be ambiguous in meaning.
• It is generally better to use a previously established and validated measure than to develop a new one for yourself
• Ratio level variables are usually counting variables or quantity measures, how many or how much. Because they have a natural zero point (none or not any), they often are "monopolar" (no meaningful opposite) and only take on positive values.
• When developing a measure of a concept that has polar opposites, such as love versus hate, use the same adjectives at each end. It will produce a more balanced measure.
• Variables always have more than one characteristic, attribute, or quality to which they refer. Constants do not.

ToTOP of the page
PROBLEM FORMULATION

• Every research question can be stated as an hypothesis, and vice versa. For example, the research question "Is there a gender difference in sentencing outcomes?" can be restated as an hypothesis, "There is a gender difference in sentencing outcomes." When converting a question into an hypothesis, one must be careful to use "one-tailed" hypothesis when the question implies the nature or direction of the relationship and a "two-tailed" hypothesis when the question does not imply the nature or direction of the relationship.
• It is easier to discuss a problem given a single research question than to derive a research question from a discussion of a problem.
• Research problems have to be stated as questions about facts, not about moral judgment

ToTOP of the page
SAMPLING

• When sampling from a population of a known size (a finite population) the sampling error is smaller than if the sample is from a population not having a known size (an open-ended population).
• Systematic sampling was early used in Persia in its war with Greece. Captured soldiers were lined up in groups of ten. One was selected from the first group and executed. Every tenth person in the line from that start was then killed. This is the origin of decimation, the killing of every tenth person (by means of a systematic sample).
• Sampling without replacement of missing cases will generally produce more representative samples than sampling with replacing of missing cases. Even if there is a lot of missing cases, the replacement cases may not compensate for systematic bias in the dropouts.
• When sampling from a population in which a target group occurs infrequently, it is better to use a stratified sample than random or systematic samples because you are more likely to have subjects with the characteristics you desire.
• When selecting random numbers from a table, only use as many digits as you need to get the percentage that you want. For a 10% sample only use one digit because that will give you the precision you need without wasting time dealing with other digits.

TABLES

• The title of tables should indicate what variables or data are being displayed and for what sample or population.
• The sample size upon which data in a table are based should be indicated with the table.
• Don't have data in a table that you neither discuss nor use to answer the research question.

ToTOP of the page
©2011 William Holmes. WH HOME