Cover
Unit 03 • 10 Lecture Hours

Research Design and Instrumentation

Architectural frameworks, human-computer interaction measurement, NOIR scales, and sampling taxonomies.

TU Weightage: 15 - 20 MarksCore: Design Triad • NOIR Scales • Sampling Matrix • Cronbach's α
Core Memory Anchors: Research Design & Instrumentation
10-Mark Question (Design Triad or Sampling)

💡 Elaboration Strategy: परीक्षा हलमा यी मुख्य Anchor Points स्मरण राख्नुभयो भने प्रत्येक बुँदालाई प्राज्ञिक रूपमा विस्तार गरेर सजिलै २ देखि ३ पृष्ठको पूर्ण १०-मार्क्सको उत्तर तयार गर्न सकिन्छ।

1

Design Triad: Descriptive -> Relational -> Experimental

Descriptive ले 'के भइरहेको छ' मात्र देखाउँछ (जस्तै कति जनाले Bug report गरे)। Relational ले दुई कुराबीच सम्बन्ध हेर्छ (स्क्रिन टाइम बढे आँखा दुख्छ कि दुख्दैन)। तर Experimental ले मात्र Cause-and-Effect प्रमाणित गर्छ (Independent variable बदलेर Dependent variable मा के असर पर्‍यो हेर्ने)।

HCI को उदाहरण दिनुस्: button को color (IV) परिवर्तन गर्दा user को checkout speed (DV) मा के फरक पर्‍यो।
2

NOIR Scales: डेटा नाप्ने ४ तह (Nominal, Ordinal, Interval, Ratio)

Nominal ले केवल नाम/वर्ग छुट्याउँछ (Mac vs Windows); Ordinal मा rank हुन्छ तर दूरी बराबर हुँदैन (1st, 2nd, 3rd); Interval मा दूरी बराबर हुन्छ तर शून्य मनगढन्ते हुन्छ (0°C तापक्रम); Ratio मा साँचो शून्य (True Zero) हुन्छ (0 ms latency, 0 MB memory)।

Exam मा लेख्नुस्: Chi-Square गर्न Nominal चाहिन्छ, तर t-test र ANOVA गर्न कम्तीमा Interval वा Ratio चाहिन्छ।
3

Reliability (अडिकपन) vs Validity (सटीकता)

Reliability भनेको जुनसुकै बेला नापे पनि एउटै नतिजा आउनु हो (Cronbach's α ≥ 0.70)। Validity भनेको हामीले जे नाप्न खोजेको हो, साच्चिकै त्यही कुरा नापिएको छ कि छैन भन्ने हो। भरपर्दो (reliable) हुँदैमा त्यो सही (valid) हुन्छ भन्ने ग्यारेन्टी हुँदैन।

Exam मा डार्टबोर्ड (Dartboard) को चित्र बनाउनुस्: सबै काँडा एकातिर लागेको तर बिचमा नलागेको = Reliable but Invalid; ठ्याक्कै बिचमा लागेको = Reliable & Valid।
4

Sampling को वर्गीकरण: Probability vs Non-Probability

यदि हरेक मान्छे छानिने बराबर सम्भावना छ भने त्यो Probability sampling हो (Random, Systematic, Stratified, Cluster)—यसको नतिजा सिधै समाजमा लागु गर्न मिल्छ। तर साथीभाइ वा सजिलोका आधारमा छान्नुभयो भने त्यो Non-Probability हो (Convenience, Purposive, Snowball)।

Exam मा २ वटा मुख्य हाँगा (Probability vs Non-Probability) भएको tree diagram बनाएर ४-४ वटा sub-types देखाउनुस्।
Section 01

Descriptive, Relational & Experimental Research Designs

In computer applications and HCI, choosing the correct research design dictates the validity of your conclusions. The three primary designs form an ascending hierarchy of scientific control:

1. Descriptive Design

Focuses on factual representation without manipulation. Answers “What is happening?” Uses user surveys, usage logging, and demographic profiles.

Key metric: Percentages, frequencies, means.

2. Relational Design

Investigates statistical associations between variables without intervention. Answers “Are X and Y linked?” (Cannot prove causation due to third-variable bias).

Key metric: Correlation coefficient (r, ρ).

3. Experimental Design

The gold standard for proving Causality. Manipulates the Independent Variable (IV), measures the Dependent Variable (DV), and controls confounding factors.

Key requirement: Control Group & Random Assignment.
👨‍🏫शिक्षकको बोर्ड नोट (Teacher's Board Note)
गोल्डेन रुल: Correlation does not imply causation! “RAM बढी हुँदा सफ्टवेयर छिटो चल्नु” सम्बन्ध (Relational) हुन सक्छ। तर “यसैले गर्दा चलेको हो” भनेर ठोकुवा गर्न अरू सबै बाहिरी फ्याक्टर (Processor, SSD, Background apps) लाई नियन्त्रण गरेर ल्याबमा Experimental परीक्षण नै गर्नुपर्छ।
Section 02

Measuring the Human in Computer Applications & HCI

Based on Jonathan Lazar et al.'s seminal framework, evaluating human interaction with computing software requires triangulating three distinct measurement modalities:

Behavioral Measures

Objective observable performance: Task completion time, Error frequency, Keystroke logging, Eye-tracking gaze fixations, and Mouse trajectory heatmaps.

Physiological Measures

Biological nervous responses to computational stress: Galvanic Skin Response (GSR), Electroencephalogram (EEG brainwaves), Heart Rate Variability (HRV).

Subjective Measures

Self-reported user experiences using standardized psychometric instruments: System Usability Scale (SUS), NASA-TLX (cognitive workload), and Likert scales.

Section 03

NOIR Measurement Scales (Stanley Stevens Taxonomy)

Every variable collected in research belongs to one of four hierarchical levels of measurement:

Scale LevelMathematical PropertiesPermissible StatisticsComputer Applications Example
N — NominalMutually exclusive category labels only. No order.Frequencies, Percentages, Mode, Chi-Square (χ²).Programming Language: [Python, C++, Java, Rust].
O — OrdinalOrder/rank exists, but distance between ranks is unknown.Median, Percentiles, Spearman's Rank (ρ).Security Vulnerability Level: [Low, Medium, High, Critical].
I — IntervalEqual unit distances, but NO true absolute zero.Mean, Standard Deviation, t-test, ANOVA.CPU Temperature in Celsius, System Usability Score (SUS).
R — RatioEqual units AND true absolute zero (absence of quantity).All arithmetic, ratios, multiplication, division.Execution Latency (ms), Download Speed (Mbps), File Size (MB).
⚠️ Examiner Mark-Deduction Trap:Likert Scale (उदा: 1 = Strongly Disagree देखि 5 = Strongly Agree) प्राविधिक रूपमा Ordinal हो। तर थुप्रै अनुसन्धानकर्ताहरूले धेरै वटा Likert प्रश्नको औसत निकालेर त्यसलाई Interval जस्तै मानेर t-test चलाउँछन्। परीक्षामा यो विवाद (Ordinal vs Interval) उल्लेख गरे परीक्षकले Full Marks दिन्छन्।
Section 04

Reliability, Validity & Cronbach's Alpha

Reliability (विश्वसनीयता = Consistency)

The repeatability and stability of measurement. Types:

  • • Test-Retest: Administering the same test twice over time.
  • • Split-Half: Correlating odd vs even numbered items.
  • • Internal Consistency: Measured via Cronbach's Alpha (α).

Validity (वैधता = Accuracy & Truthfulness)

Ensures the test measures the actual intended theoretical construct:

  • • Content Validity: Do items cover the complete domain?
  • • Internal Validity: Did IV cause DV without extraneous bias?
  • • External Validity: Can findings be generalized to the real world?
CRONBACH'S ALPHA (α) FORMULA & INTERPRETATION:Threshold: α ≥ 0.70
α = [ k / (k - 1) ] × [ 1 - (Σ sᵢ² / s_total²) ]
(where k = number of survey items, sᵢ² = variance of item i, s_total² = total score variance)
α < 0.60: Unacceptable0.70 ≤ α < 0.80: Acceptableα ≥ 0.80: Good / Excellent
Section 05

Master Sampling Taxonomy: Probability vs Non-Probability

Sampling defines the process of selecting a representative subset from the Target Population:

PROBABILITY SAMPLING

Generalizable
  • 1. Simple Random: Every item has an equal non-zero probability (Random Number Generator / Lottery).
  • 2. Systematic: Sampling every k-th element from an ordered list where interval k = Population N / Sample n.
  • 3. Stratified: Dividing population into homogeneous strata (e.g., Year 1 vs Year 2 MCA scholars) and sampling proportionally.
  • 4. Cluster: Dividing heterogeneous population into geographic or organizational clusters and selecting entire clusters.

NON-PROBABILITY SAMPLING

Exploratory / HCI
  • 1. Convenience: Selecting subjects who are easiest to reach (e.g., campus lab students).
  • 2. Purposive / Judgmental: Hand-picking subjects based on expert criteria (e.g., Senior DevOps engineers).
  • 3. Quota: Selecting non-random quotas to match demographic ratios without random generation.
  • 4. Snowball: Existing subjects recruit their acquaintances (essential for hidden populations like ethical hackers).
Section 06

Qualitative Methods & Automated Data Collection in CA

Qualitative Methods in Computer Applications

  • • Diaries: Participants log their daily software friction and bug encounters.
  • • Case Studies: In-depth multi-faceted inquiry into a single enterprise system migration.
  • • Focus Groups: Moderated discussions (6-10 users) exploring UI aesthetic impressions.
  • • Ethnography: Observing developers in their natural workplace setting without intervention.

Automated Data Collection Methods

  • • Web Scraping & APIs: Structured extraction using Python BeautifulSoup/Scrapy.
  • • System Telemetry & Server Logs: Apache/Nginx request logs, crash stack traces.
  • • Clickstream Tracking: Real-time user event streaming via JavaScript instrumentation.
  • • IoT & Sensor Streams: Environmental and biometric time-series monitoring.
High-Probability Exam Blueprint

TU Exam Model Q&A Bank — Unit 3

5 Past & Model Questions

💡 यी प्रश्नहरू विगतका TU परीक्षा तथा सिलेबसका मुख्य उद्देष्यहरूबाट संकलन गरिएका हुन्। उत्तरपुस्तिकामा Full Marks पाउन चाहिने मोडेल ढाँचा यहाँ प्रस्तुत गरिएको छ।

शिक्षकको छोटो सारांश (One-Minute Exam Recall):
रिसर्च डिजाइन भनेको घर बनाउनुअघि तयार पारिने 'इन्जिनियरिङ नक्सा (Blueprint)' जस्तै हो। यसले कुन डाटा कसरी संकलन गर्ने, कुन स्केल प्रयोग गर्ने र कसरी विश्लेषण गर्ने भन्ने सम्पूर्ण योजना निर्धारण गर्छ।

Standard Model Answer Structure:

1. Definition: A Research Design is the master architectural blueprint and conceptual framework within which an empirical investigation is conducted. It outlines procedures for data collection, measurement, and statistical analysis.

2. Essential Features: Objectivity (minimizing researcher subjectivity), Reliability (consistent observational outcomes), Validity (accurate measurement of intended variables), Generalizability (applicability to wider populations), and Resource Economy (achieving high accuracy within realistic budgetary and hardware constraints).

✍️ Essential Answer Keywords / Must-Draw Elements:
✓ Master Blueprint✓ Objectivity & Economy✓ Control of Extraneous Variables✓ Sampling & Measurement Plan
Previous: Unit 2Next: Unit 4 (Statistical Tests & Numericals)