Research Approach

Wither descriptive or explanatory research needs to have a frame of reference to interpret results
within and not just report results.

Descriptive research
The need for a frame of reference in descriptive research is fairly obvious, for example,
unemployment rate of an area is 15%, but is this high or low, is it increasing or decreasing over
time, how it compares with other areas or of other years, etc.
There is a need to provide a context to make sense of data and to design methods to ensure that
needed data to provide this context are collected.
HOW to do this? Collect data about other groups and about the same group over time.

Explanatory research
Deals with causal processes, the goal is to develop an explanation of the patterns in the data and
need to eliminate as many alternative explanations of the patterns as possible.

1. Classic experimental design

An experimental design to measure the effect of distributing flyers on building towers preference in
residential areas in Amman

T1 (before) T2 (after) Deff = T2 - T1

E1 Intervention E2
Experimental 40% support 33% support Ediff = 33-40
Read flyers = -7%

C1 C2 C diff = 39-40
Control 40% support 39 support = -1%

If change were significant we would conclude it is because of the experiential intervention.

Example characteristics of a control group to achieve an acceptable conclusion are:

 Same variable to start with (e.g. a group, neighbourhood, a city, etc.)

 Had same identical experiences between time one and time two.
 To ensure similarity sample is assigned randomly to the experimental and control groups
Large sample helps similarity in most respects, where ensuring same experiences between time one
and time two is more difficult particularly in social research than in labs for a number of problems
University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 1
Scientific Research methodology
 such as not possible to get repeated measures for the same group, thus impossible to obtain
measures at both times 1 and 2 and often difficult to do a control group.
2. Panel Design

This design uses only the top two cells of the experimental design; it looks at the same group over a
period of time.

3. Quasi– panel design

4. Retrospective panel design

5. Retrospective experimental design

6. Cross - sectional or correlational design

One – shot case study

Sampling
Sampling provides an efficient and accurate way of obtaining information about a large number of
cases. How efficient and accurate depends on the type of sample used, size of sample and the method of
collecting data from the sample. In the end the decision about samples will be a compromise between
cost, accuracy, the nature of research problem and the art of possible.

There are tow types of sampling: probability and non-probability sampling

Probability
Simple random sampling, systematic sampling, stratified sampling, multistage sampling.

Sample size
Depends on two key factors:
 degree of accuracy for sample, and
 the extent to which there is variation in the population
There is a need to decide on how much error we are prepared to tolerate

Non response
 Sample size reduction (eg. add 20%) &
 Bias (old age, migrants, lower education, etc.) what is the bias and its extent (info from
observation, define characteristics of non respondents and adjust accordingly, then compare
with others characteristics from sample).

2. Non- probability sampling

Purposive sampling: typical of a category useful to the researcher

Quota sampling: interviews to find types of particular characteristics then given quotas. It is non
random & causes bias and difficult to estimate accuracy of a sample.

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 2

Scientific Research methodology
Constructing Questionnaires

Selecting fields of questions

● Think ahead of needed information to inshore that relevant questions are asked, the questionnaire
should reflect both theoretical thinking and an understanding of data analysis.
● Ways to find which questions to ask are:
1. What concepts to measure according to the research problem
2. Indicators devised for the concept are important in determining questions to ask.
3. Researcher expectation about how variables are linked and factors explaining
relationships requires including a certain kind of questions.
4. The method of data analysis affects needed information to collect.
● Four aspects of the questionnaire in explanatory research to think of:
1. Measures of the dependent variable(s): clarify what to explain & develop questions to
measure it.
2. Measures of the independent variable(s): make questions to tap the causal variables.
3. Measures of test variables: variables which helps to clarify the nature of links between
independent and dependent variables.
4. Background measures: questions about characteristics such as age, religion, education etc.
This information helps see if patterns differ for various subgroups.
● Question content: distinguish between 4 types of question: behavior, beliefs, attitudes & attributes
1. Behavior (what people do).
2. Beliefs what people think is true rather than accuracy of their beliefs.
3. Attitude questions try to establish what respondents think is desirable. (whether or not, e.g.
whether mothers should work or not).
4. Attribute questions are designed to obtain info about respondent’s characteristics (age,
education, ethnicity, etc).
Type of needed information, awareness of the 4 types should lead to a systematic development of
questions to tap all types of required data. In the analysis, in the development of scales its important
to develop composite measures normally of the same type of information. It is inappropriate to
combine the 4 types of info into a single measure because they tape different things.

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 3

Scientific Research methodology
Wording questions
Clear and simple language, short, not double questions, not leading nor negative, consider respondent
knowledge, do words have same meaning to everyone, prestige bias (what does an important person to
the respondent think about something), ambiguity (words with several meanings), when to ask direct or
indirect questions particularly in sensitive areas. Put a clear frame for question (e.g. how often do u do
this, is it in one week or one month, etc.).
The question should not artificially create opinions, respondents may not have an opinion of an issue,
questionnaire should include options ‘don’t know’ or ‘no opinion’, this may lead respondents to give
this response to most questions, but not including it will produce highly unreliable & useless responses.
Is personal wording (how they feel) or impersonal wording (how people feel) is preferable, this depends
on what to do with the answer, the impersonal provides a measure of respondents perception of other
people’s attitudes not his. Does a question wording need to be detailed (put in a category e.g. an age or
income groups. Put subject mater first before giving alternatives.
Does the question produce a response set, if respondents are asked to agree or disagrees they may
simply agree regardless on their true opinion, this can produce a set of acquiescent response set (if this
appear try to make the respondent feel comfortable regardless of the answer they may give, and avoid
making a certain answer appear normal or unusual).

Selecting question type

Open and closed formats, a suggested combination is:
1. a closed question to see if the respondent has thought about or is aware of the issue;
2. an open question to know about the general feeling on the matter
3. a closed question to get at specific aspects of the issue;
4. apen or closed question to find out respondents’ reasons for their opinions;
5. closed question to find out how strongly the opinion is held.

Types of forced-choice response formats: rating scales

1. Likert-style formats
1. verbal: strongly agree, agree, cannot decide, disagree, strongly disagree
2. Diagrammatic - 0 +
Strongly disagree strongly agree
1 2 3 4 5 6 7 8 9 10

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 4

Scientific Research methodology
2. Semantic differential formats
e.g. how would you describe your street? To ask respondent to put a mark between two extremes
Busy 1 2 3 4 5 6 7 not busy
Noisy 1 2 3 4 5 6 7 Quite

3. Checklists, consists of a list of items and respondents circle relevant items.

4. Ranking formats: to give a list of alternatives and respondents to rank their importance.
5. Attitude choices not agree-disagree statements: which option do u favor?
{ } City should expand to the West
{ } City should expand to the East
{ } No expansion of the city to West or East
{ } Cannot choose
6. Direction extremity and intensity of attitudes, important to distinguish the direction of a
person’s attitude from the extremity of his position and intensity with which they hold that
position. We can know direction by asking to chose between two alternatives. To learn more by
asking how extreme the respondents view is: 1. give a statement and respondent say how
strongly they agree or disagree, or 2. by asking respondent to put a mark between two extremes.
A follow up question can be to know how strongly respondent hold his position to measure his
attitude intensity.
7. Vignettes, a short paragraph followed by a number of questions.

Questionnaire layout: answering procedures, contingency, instruction, use of space, order of

questions, coding.
Telephone and mail questionnaires
Pilot testing, evaluating questions and questionnaire
Administering questionnaire

Analysing Data

Three factors affecting how data are analysed:

1. the number of variables being examined
2. the level of measurement of the variables
3. the use of data whether for descriptive or inferential purposes.

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 5

Scientific Research methodology
The number of variables
One variable- univariate method (e.g. age, income, etc.), two variables bivariate method (e.g. if income
and house condition are related), three or more variables multivariate technique (e.g. we might be
interested in housing conditions: why some houses better than others, income level, family size,
education, etc.). before analyzing we need to be clear about the question we are trying to answer. In
practice research questions are refined in the process of analysis so there could be movements between
univariate, bivariate and multivariate.

Levels of measurement

There are three main levels of measurement: nominal, ordinal and interval/ratio
1. Nominal level variables: can be distinguished but not ranked (country, religion, sex, marital status)
2. Ordinal variable: its meaningful to rank the categories, but not possible too quantify difference
between categories (e.g. paid and unpaid job, part time or full time) categories can be ranked but
differences between categories cannot be quantified in numerical terms is an ordinal variable.
3. An interval/ratio variable: categories can be ranked and it is possible to quantify precisely the
differences between categories (e.g. income, if measured in money, is an interval because as well as
ranking respondents according to their income, the precise difference between income can be
quantified.
Level of measurement
Nominal Ordinal Interval
Are there difference categories: yes yes yes
Can categories be ranked: no yes yes
Can differences between
categories be specify numerically no no yes

Descriptive and inferential statistics

Descriptive statistics: summaries patterns in the responses of people in a sample (info about income,
education, age, etc).
Inferential statistics: provide an idea about whether the patterns described in the sample are likely to
apply in the population from which the sample is drawn.

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 6

Scientific Research methodology
1. Univariate analysis
2. Bivariate analysis
3. Multivariate analysis

1. Univariate analysis

1. Descriptive statistics used for nominal, ordinal and interval variables

2. Inferential statistics interval estimation statistics

Descriptive statistics

1. Nominal variables:
Frequency distributions

Table ( ) Respondents’ relations with their neighbors in building1 and building2

Neighbors Relatives Visits in the past 5 years
% % %
building1 (b1) 20 16 9
building2 (b2) 17 7 3

n (b1)= 100, n* = 2, n (b2)= 31, n* = 1 Source: author’s survey 2001

n (number of respondents), n* (number of non response)
note: non response is not included in percentages

Histograms (charts)

N = 200

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 7

Scientific Research methodology
 Figure ( ) Respondents’ opinions in building1 and building2
about wishing to stay in their home or not
n (building1) = 100, n (building 2) =31, Source: author’s survey 2020

Central tendency (the mode) is the most common response

e.g. Frequency table for age groups (population over 15 years old)

16-20 20%
21-29 55%
30-39 5%
40-49 12%
50-60 7%
Total 100%
The mode is the age group 21-29

It is not typical, it might be typical for a group of 100 but not as a whole, some distributions have more
than one mode, vulnerable to how categories are combined, thus mode is unstable and open to
manipulation.

Dispersion: the variation ratio (v) ‫التشتت‬

v : percentage of people who are not in the modal category, the higher the percentage the more poorly
the mode reflects the overall distribution, and expressed as a proportion. It is expressed as a portion. To
calculate (v) move decimal place 2 places to the left, e.g. if 60 per cent of respondents are not in the
modal category, then v = 0.06

2. Ordinal variables

Frequency distributions

Categories of ordinal variables are ranked thus should be put in a rank order in a frequency table.

Frequency of visits of neighbours

N Per cent Cumulative %
Daily 50 20 20
Several times a week 300 30 50
Weekly 150 15 65
2-3 weeks 100 10 75
Monthly 150 15 90
University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 8
Scientific Research methodology
Less often 100 10 100
Total N 1000 100%

The cumulative percentage is a rolling addition of each of the percentages in earlier categories of the
variable. Thus, the cumulative percentage of 65 per cent means that 65 per cent visit their nighbours
weekly or more often.

Central tendency: the median (middle) rank from low to high

age 12 14 16 18 20 22 24

Case 1 2 3 4 5 6 7 8 9 10
class KG1 KG2 1st 2nd 3ed 4th 5th 6th 7th 8th
median

Dispersion: the decile range

If most cases in a distribution are in ranked categories close to the median category, the median is a
good summary of the group, if many cases are long way from median category its not so good.

1.
Case 1 2 3 4 5 6 7 8 9 10
no. of men 0 1 2 2 2 2 2 2 3 10
median
Median =2 Extreme score, high range underestimates the usefulness of the median
Range =10
Decile range = (3-1) = 2

2.
no. of men 0 1 1 2 2 2 2 2 2 3
median
Median=2 Low range reflects adequacy of median
Range = 3
Decile range = (2-1) =1

3.
no. of men 0 1 2 2 2 2 5 6 9 11
median
Median=2 High range accurately shows inadequacy of median
Range = 11
Decile range = (9-1) = 8

To avoid distorting effect of extreme cases, drop bottom 10 per cent of cases (first decile) and the
top 10 per cent and look at the middle 80 percent
University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 9
Scientific Research methodology
Interval variables
Variables which categories can be ranked and differences between categories can be quantified in
precise numerical amounts.
Frequency
Tables for interval variables are similar to ordinal variables. If interval variables have a large
number of values (e.g. age, income $, etc.) its better to group values (e.g. 10-19, 20-29, etc.). note
that its desirable to make all categories of similar width & have open ended category (e.g. 10-19,
20-29, 30 plus).
Central tendency (the mean X )
Its calculated by adding up scores for each case in the sample and divided by N, the member of
cases in that sample.

Case $ Income
1 12 000
2 13 000
3 15 000 Total income = 189000
4 16 000 Total cases N = 10
5 18 000
6 20 000 X = 18900
7 21 000
8 22 000
9 25 000
10 27 000 Main problem is it can be distorted by extreme cases, e.g. if a person earns
$1 million was added to the table below mean would be $108 091, thus mean doesn’t reflect
adequately reflect the bulk of the group. Another problem is its possible to obtain same mean for
two different distributions as:

Same mean for two different distributions

Group 1 Group 2
Age N Age N
30 0 30 40 X Group 1
35 10 35 10 X = (30x0+ 35x10+ 40x20+45x40+50x20+55x10+60x0)
40 20 40 0 = 4500 / 100 = 45
45 40 45 0
50 20 50 0 X Group 2
55 10 55 10 X = (30x40+ 35x10+ 0x0+45x0+50x0+55x10+60x40)
60 0 60 40 = 4500 / 100 = 45
N = 100 N = 100
X = 4500/100 = 45 X = 4500/100= 45
s = 5.5 s = 14

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 10

Scientific Research methodology
Its necessary to use an appropriate measure of dispersion which is variance or standard deviation.

Dispersion: variance s & standard deviation (s)

2

1. Provides a measure of the summarizing value of a mean and tells us within what range of the
mean a given percentage of cases lies, i.e. to see how far each case is from the mean, then add up all
the deviations and obtain an overall average of these deviations to use as the measure for dispersion.
Variance s 2 = ∑ (Xi – X)2 Standard deviation s = √ s2
N
Xi = an individual’s score on the variable (e.g. age)
X = the mean, (e.g. 45)
N = total number of cases in the sample.

Standard deviation (s 2 ) Group 1

Age N Xi-X (Xi-X)2 N (Xi – X)2
30 0 -- -- --
35 10 -10 100 1000
40 20 -5 25 500
45 40 0 0 0
50 20 5 25 500
55 10 10 100 1000
60 0 0 -- --
Total N = 100 Total = 3000
X = 45 s 2 = 3000 = 30 s ≈ 5.5
100

The lower s the better the mean is as a summary measure. The standard deviation (s) for Group1
(=5.5) is more satisfactory than s for Group 2 (=14). In other words s shows that the mean is a
more accurate summary for group 1 than for group 2.
From probability theory, it is known that in a normal distribution it is true that 68 % of cases will lie
within one standard deviations about or below the mean. In this case the size of a standard deviation
is 5.5, that is 5.5 years. Thus, 68% of cases will be within a range of 45 years (mean) plus or minus
5.5 years, i.e. 39.5 to 50.5 years. Probability theory tells us that 95 % of cases will always lie within
plus or minus two standard deviations of the mean, in this case 11 years of the mean (34-56 years).
When a distribution is not normal, the percentage of cases which lie within various numbers of
standard deviations of the mean cannot be predicted with as much precision as summarized in the
below table (from probability theory):

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 11

Scientific Research methodology
Per cent within various standard deviations for various types of distribution
Number of Standard Minimum for Minimum for any unimodal Value for a normal
deviations from X any distribution & symmetrical distribution distribution
1.00 0 56 68
2.00 75 89 95
3.00 89 95 99.7

2. Standard deviation has another use; any score on a variable can be converted into a standard
score called (z- score), it’s a particular score expressed as a standard deviation units. e.g. if a mean
age is 50 years and s = 5, the age 60 is expressed with a z-score of 2 (mean+2s, i.e. 50 +10).
To convert any score into a z-score use the formula z = (Xi – X)
s

e.g. if age is 53, z = 53-50 = 0.6, for the age 45 z-score= -1(mean - 1 standard deviation)
5

Inferential statistics

Used to help explore from the patterns in a sample to likely patterns in the population of the sample.
There are two technique:
1. Interval estimates most common used in univariate analysis (using the mean, ordinal data)
2. Inference for non-interval variables (mean cannot be calculated, i.e. nominal or ordinal data)

1. Interval estimates
Samples are always liable to some error, it is crucial when trying to generalize from sample
estimates to use the relevant inferential statistics. With univariate analysis the most widely used
technique in survey research is to use the standard error which enables us to estimate the population
patterns within a range. This procedure is called interval estimation.

If in a sample the mean income is $18000, what is the mean for the population. Since samples are
unlikely to be a perfect reflection of the population (sampling error), we can’t use the sample mean
(sample estimate) to suggest the actual mean income for the population (population parameter).

We need a way of estimating how accurate our sample estimate is likely to be and how close is it
likely to be the population mean. Standard error of the mean formula is Sm = s
Sm = Standard error of the mean , s = standard deviation N total no. in sample √N

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 12

Scientific Research methodology
 A distribution of random sample means

No. of samples True population mean

60
50 Distribution of the means of a large
40 number of samples
30
20
0
12 14 16 18 20 22 24 Mean income ($1000)

Probability theory says that for 95% of samples the population mean will be within ¯+ 2 standard
error units of the sample mean, i.e. in our sample there is a 95% chance that the population mean
will be within ¯+ 2 standard errors of the population mean. We can then estimate within a range
where the population mean is likely to be. This range is called the confidence interval and our
degree of certainty that the population mean will fall within that range. 95% is called the confidence
level. The units of Sm are the same of the variable concerned, e.g. income in $ or age in years.
What does this mean in our language? For example, if a sample mean is $18,000 and Standard
error (Sm) is $1,000. Thus there is a 95% chance that the population mean is within the range
$16,000-$20,000 (i.e. ¯+ 2 standard error). The size of S m is a function of sample size, i.e. the
smaller Sm the larger the sample size. Quadrupling the sample size halves Sm.

Inference for non-interval variables

If mean cannot be calculated (i.e. nominal or ordinal data) first we have to variable must have only
2 categories; the formula is SB = PQ ,
√ N
SB = standard error for binominal distribution
P = % in on category of the variable
Q = % in the other category of the variable
N = total number in the sample.

Look at the percentage of respondents answering a question in a particular way and ask how close
to it the percentage in the population is likely to be. E.g. sample of 1000, 52% intend to vote labor
(thus 48% will vote other way), estimate how the population will vote?

To calculate the standard error SB = 52 x 48 = √ 2.49 = 1.6 what does it mean?

√ 1000

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 13

Scientific Research methodology
There is 95% chance that the percentage in the population who favor Labour will be ¯+ 2 standard
error of the sample percentage, i.e. 52% ¯+ 3.2 % (¯+ 2 standard errors) between 48.8% & 55.2%.

In summary, samples are always liable to some error, it is crucial when trying to generalize from
sample estimates to use the relevant inferential statistics. With univariate analysis the most widely
used technique in survey research is to use the standard error which enables us to estimate the
population patterns within a range. This procedure is called interval estimation.

Summary
A range of descriptive and inferential statistics for univariate analysis have been outlined. The
statistics chosen depend on the level of measurement of the variables and these are summarized in
the table below:

Selecting statistics for univariate analysis

Descriptive statistics inferential statistics
Level of measurement Central tendency Dispersion
Interval estimate using
Nominal Mode Variation ratio
standard error of binominal
Ordinal Median Decile range As above
Mean Standard deviation Interval estimate using the
Interval
or variance standard error of the mean

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 14

Scientific Research methodology
 2 Bivariate Analysis: crosstabulations

Surveys In Social Research

David De Vaus

The main text of the social science research report or thesis consists of 5 chapters

1. Introduction: This includes a brief introduction of the topic or research problem,

significance of the topic, and relevance of the topic in the current scenario.
2. Literature review: This includes all the references that you have done to do this
investigation.
3. Methodology: The methodology chapter must include the objective of the study,
variables, limitation of the study, data collection tools, research design, sampling
methods, the population and sample details etc.
4. Data analysis and Interpretation: in this chapter, you must include all the tables,
charts, graphs related to the collected data. You must provide an analysis of each
data provided.
5. Findings and conclusion: In this chapter, you can briefly explain your
interpretation and results of your research. You can also compare and contrast your
findings with the help of a theory. You can provide suggestions also.

University of Jordan, Dept. of Architecture Dr Firas Sharaf 2020 © 15

Scientific Research methodology