Research Methodology final assignment

Name Registration number

Eric Mugo SCT 213-C002-0117/2021
Joseph Abuya SCT213-C002- 0112/2022
Faith Gichuru SCT213-C002- 0003/2022
Amon Tot SCT213-C002- 0060/2022
Jarvan Owuor SCT213-C002- 0022/2022
Dennis Guchu SCT213-C002- 0006/2022
Levi Omondi SCT213-C002- 0107/2022
Ashley Otieno SCT213-C002- 0070/2022
Zawadi Mukuru SCT213-C002- 0064/2022
Alexander Kihoi SCT213-C002- 0002/2022

i)Critically evaluate the philosophical underpinnings of two contrasting research paradigms (e.g.,
Positivism and Interpretivism) and discuss how each paradigm would shape the research design,
data collection methods, and interpretation of findings in a study investigating the societal
impact of explainable AI (XAI) in financial credit scoring in Kenya. (10 marks)

Positivism is a school of thought whose assumption of reality is that it’s objective and that the world
can be observed and measured. Knowledge is generated through empirical observations or
measurements and using hypothesis testing in the scientific method. A positivist researcher aims to be
objective and that their work is replicable.

Constructivism on the other hand is where we say that reality is socially constructed and subjective,
and the focus of constructivism is to understand the world through the eyes of the participant.
Knowledge in the constructive framework is generated through interpreting people’s subjective
experiences of a situation. The researcher in this setup has to acknowledge their own role and
interpretations while taking part in a research study which is constructive in nature.

We can say in the positivist framework that the most appropriate research design would be a
quantitative one, here the focus is on Exploratory Data Analysis, Experimental Designs for experimental
data or Causal Designs for observational data. The focus here is on generating knowledge through
statistical analysis in order to generate insights from the objective reality.

In a constructivist framework the most appropriate research design would have to be a qualitative
one. Here we focus on questionnaires (either open or closed ended), surveys, interviews, etc. We can
also use case studies. The responses generated will be inherently subjective so they will be subject to
interpretation.

The positivist framework’s objective of quantifying the social impact on credit scoring will involve
either an experimental or a quasi-experimental design where we test a given hypothesis. Here an
example of a null hypothesis would be about how there would be no significant difference in the
average loan approval rates of the AI model and the default model. The findings of such a study would
be generalisable where we seek the cause-effect relationships between variables.

The constructive framework’s objective would be to assess how different stakeholders feel about the
effect of AI where the qualitative design would be flexible and exploratory. Here we can use interviews,
open ended questionnaires and focus groups to gather this subjective data. Here we get a deeper
understanding of the participants’ sentiments and where they come from.

(ii) Differentiate between probability and non-probability sampling techniques, and in the
context of researching user acceptance of a new mobile-based health monitoring application in a
diverse urban setting like Nairobi, propose and justify the most appropriate sampling strategy.
Discuss potential biases introduced by the chosen method and how they might be mitigated. (10
marks)

Probability Sampling Techniques

Probability sampling methods involve random selection, ensuring that every individual in the
population has a known and equal chance of being selected. This helps in obtaining a representative
sample that mirrors the diversity of the population.

One such method is simple random sampling, where each individual has an equal chance of being
chosen. For a research study on user acceptance of a mobile-based health monitoring application in
Nairobi, this method would involve randomly selecting users from a complete list of potential
participants. If the target population is not easily identifiable, this could be challenging, as obtaining an
accurate list may be difficult, especially if the app is being introduced to a broad, undefined user base.

Another approach is stratified random sampling, which divides the population into distinct
subgroups or strata, such as age, income levels, or geographical location. From these subgroups,
random samples are then taken. In the case of Nairobi, stratified sampling would ensure that various
demographic groups, like different age categories or people living in different neighbourhoods, are
adequately represented. This would allow researchers to understand how each demographic group
views the health monitoring app and could highlight variations in its acceptance across different
sections of the city.

Non-Probability Sampling Techniques

Non-probability sampling does not rely on random selection, and as such, it is less likely to produce a
sample that is representative of the broader population. However, these methods can still be useful in
certain situations.

Convenience sampling is one non-probability technique where participants are chosen based on their
accessibility or willingness to participate. In this context, it would involve selecting users who are easy
to reach, such as those already using the app or those in locations where recruitment is easier.
However, this method can introduce bias since the sample may not reflect the broader population. For
instance, it may over-represent tech-savvy users or those with easier access to smartphones, which
could skew the findings.

Snowball sampling is another method where initial participants are asked to refer others who meet
the study criteria. This technique can be helpful if the target population is difficult to identify or if
certain groups (like early adopters of health technology) are more likely to know others with similar
characteristics. While this method can help researchers reach hard-to-find participants, it has a
downside: it may introduce bias because the sample is built on personal networks and thus may not be
diverse.

Proposed Sampling Strategy

Given the context of Nairobi, a stratified random sampling approach would be the most appropriate
for studying user acceptance of the mobile-based health monitoring application. Nairobi is a highly
diverse city with variations in socioeconomic status, access to technology, and healthcare needs.
Stratified sampling would ensure that different segments of the population are represented in the
sample, providing a more comprehensive picture of how different groups might respond to the app.

For example, you could divide the population into strata based on factors such as income, age,
education, and access to technology. This would ensure that individuals from high-income areas,
middle-income areas, and low-income areas are all included in the sample. Such an approach helps
capture a variety of perspectives on the app's usability, benefits, and potential barriers to adoption.
Additionally, stratifying by demographic characteristics such as age would allow researchers to assess
whether certain age groups are more likely to accept or adopt the app.

Potential Biases and How to Mitigate Them

While stratified sampling can provide a representative sample, there are potential biases that could
affect the results. One key issue is sampling bias, where certain groups within the strata may be
underrepresented or overrepresented. To mitigate this, it's important to use accurate and up-to-date
data to define the strata, such as census data or reliable demographic surveys, ensuring that each
subgroup is represented proportionally.

Another concern is response bias, where participants may not provide truthful or accurate answers,
particularly if they feel pressure to respond in a socially acceptable way. In this case, it's important to
ensure participants feel comfortable by guaranteeing anonymity and confidentiality. Clear
communication about the research's purpose and potential benefits could help reduce this type of
bias.

Non-response bias can also arise if a large proportion of the selected participants fail to respond. This
can be addressed by following up with participants through multiple channels, ensuring that those
who do not initially respond have an opportunity to participate.

Finally, social desirability bias may occur if participants give responses, they think are more
acceptable or desirable to the researcher. To combat this, it's important to frame questions in a neutral
way and stress the importance of honest feedback to help improve the application.

Conclusion

In summary, stratified random sampling is the most suitable method for studying the user acceptance
of a mobile-based health monitoring application in Nairobi. By ensuring that diverse groups are
represented, this approach allows for a deeper understanding of how different segments of the
population might react to the app. However, the chosen method could introduce biases, such as
sampling, response, and non-response biases. These can be mitigated by careful planning, ensuring
accurate representation, and employing strategies to reduce bias, ultimately leading to more reliable
and valid research findings.

(iii) Consider a hypothetical longitudinal study tracking the evolution of cyber-attack patterns on
critical infrastructure in East Africa over a five-year period.

a) Formulate a complex research question that this study would aim to answer. (3 marks)

How do cyber-attacks on important systems such as banks in East Africa change over the past five
years and what causes these changes?
b) Propose a suitable research design, outlining the key phases and data sources. (4 marks)

Phase one: Purpose statements and Formulation of Objectives and Data Collection: Explore the
objective and feasibility of the research. Data sources can be past reports on cyber-attacks across the
region.

Phase Two: Data Pre-processing, Techniques and Methodology: Track attacks, consolidate them, begin
data analysis and choose the kind of technique to use, qualitative or quantitative. Data Sources can be
live attack logs and surveys.

Phase Three: Data Analysis: Use of Quantitative analysis of Time series method to look at trend of how
often attacks happen.

Phase Four: Research reports and conclusion: Compile reports gotten from the analysis of data, and
draw conclusions on what is the trend of attacks, suggest better policies to strengthen security of
systems.

c) Identify and discuss three significant ethical and logistical challenges specific to conducting
such a long-term, sensitive study. (3 marks)

1: Data safety: Ensuring data confidentiality from unauthorized access.

2: Data Accessibility: Integrity of data may be weakened by hesitation to share data for analysis or
interference of data because of change in staff/policies/management.

3: Data Accuracy: Due to the long period data may become redundant or inaccurate due to change in
management/inaccessibility/policies.

QUESTION TWO (20 Marks)

a) Evaluate the suitability of different research designs (e.g., experimental, quasi-experimental,
correlational) for investigating the causal impact of a government's new data-driven agricultural
extension program on smallholder farmer productivity in a specific region of Kenya. Justify your
choice of the most appropriate design, discussing the potential challenges and limitations in
implementing it in a real-world setting. (8 Marks)

Experimental Design (Randomized Controlled Trial - RCT)

Suitability: RCTs are the gold standard for establishing causality. Farmers would be randomly
assigned to either a treatment group (receiving the program) or a control group (not receiving the
program).
Justification: Randomization ensures that any observed differences in productivity are directly
attributable to the program. This minimizes bias and confounding variables.
Challenges & Limitations: Practical constraints may arise, including ethical concerns (denying access
to beneficial services), logistical difficulties in implementation, and possible contamination between
groups.
Quasi-Experimental Design
Suitability: If randomization is infeasible, quasi-experimental methods—such as difference-in-
differences (DiD) or propensity score matching (PSM)—can be used.
Justification: These methods allow for causal inference by comparing farmers before and after the
program or matching similar farmers who did and did not receive the program.
Challenges & Limitations: Selection bias may still exist, and finding comparable control groups might
be difficult. Additionally, external factors influencing productivity may complicate results.
Correlational Design
Suitability: Correlational studies examine associations between the program and farmer productivity
without inferring causality.
Justification: While useful for preliminary insights, correlational analyses cannot definitively establish
causality due to confounding factors.
Challenges & Limitations: Correlation does not imply causation, and unobserved variables may
distort findings.
an experimental design (RCT) would be ideal if feasible. However, due to practical and ethical
challenges, a quasi-experimental approach (such as DiD or PSM) might be more appropriate in a real-
world setting, balancing rigor with logistical feasibility.

b) Differentiate between cross-sectional and longitudinal data collection approaches. Discuss

how a mixed-methods approach, combining both quantitative and qualitative data collection
techniques, could provide a more comprehensive understanding of the factors influencing
technology adoption among rural communities in Kenya, using the agricultural extension
program scenario from (a) as an example. (7 Marks)

Cross-Sectional Approach

Collects data at a single point in time from different individuals or groups.

Provides a snapshot of the situation at that moment.

Useful for identifying patterns, trends, or associations but cannot establish causality.

Example: Surveying rural farmers in Kenya once to assess their current technology adoption levels.

Longitudinal Approach

Collects data over multiple time periods from the same subjects.

Tracks changes and developments over time.

Can help establish cause-and-effect relationships.

Example: Tracking the same group of farmers for 5 years to observe how their technology adoption
evolves.

Mixed-Methods Approach

A mixed-methods approach combines quantitative (numerical) and qualitative (descriptive)

techniques, offering deeper insights into technology adoption among rural Kenyan farmers.

Application in Agricultural Extension Programs:

Quantitative Data (Cross-Sectional or Longitudinal)

Surveys/Questionnaires: Measure adoption rates, frequency of technology use, and demographic

factors.

Longitudinal Tracking: Assess changes in adoption over time (e.g., before and after training).
Qualitative Data
Interviews/Focus Groups: Explore farmers' perceptions, barriers (e.g., cost, lack of knowledge), and
cultural influences.

Case Studies: Examine successful vs. unsuccessful adoption cases in depth.

Benefits of Mixed-Methods in This Scenario:

Comprehensive Insights: Quantitative data shows how many farmers adopt technology, while
qualitative data explains why some resist.

Triangulation: Cross-validates findings (e.g., if surveys show low adoption, interviews reveal
underlying reasons like distrust of new tech).

Policy & Program Improvement: Helps design targeted interventions (e.g., training for hesitant
farmers, subsidies for costly tools).

c) Considering the research on technology adoption in rural communities, discuss the ethical
considerations related to data privacy, informed consent, and potential digital divide biases
when collecting data from vulnerable populations. Propose strategies to mitigate these ethical
risks. (5 Marks)

1. Data Privacy:

People in the rural areas may lack awareness of how their data is collected, stored, or used, making
them feel like their privacy have been violated. Sensitive information such as financial status, health
data, or personal beliefs could easily be misused if not properly protected.

Risk - Unauthorized access or misuse of personal data due to inadequate data protection practices.

Strategy - Ensure strict data encryption, anonymization, and secure storage protocols. Researchers
should comply with data protection laws (e.g., Kenya’s Data Protection Act, 2019).

2. Informed Consent:

Low literacy levels or unfamiliarity with digital technologies can hinder true informed consent.
Participants may not fully understand the purpose or implications of the research. They may even be
skeptical in giving out their data.

Risk - Participants may be misled or coerced into participating without full comprehension.

Strategy - Use simple and culturally appropriate consent materials. You can employ verbal
explanations or visual aids, and allow time for questions where they don’t understand. Use witnessed
verbal consent when literacy is a barrier.

3. Digital Divide Bias:

The research may unintentionally favor some individuals with better access to technology, excluding
those without smartphones or internet, thus skewing data and reinforcing systemic inequalities.

Risk - Underrepresentation of the most vulnerable populations and misinformed policy decisions.

Strategy - Use inclusive sampling methods such as offline surveys, paper-based tools, or community
interviews. Provide access to devices or internet during data collection sessions.