(a) Critically examine the top-down and bottom-up processing in perception. How can we make the combined use of both the…

(a) Top-down and bottom-up processing in perception Perception involves two interacting mechanisms: bottom-up (data-driven) and top-down (conceptually driven) processing. Bottom-up processing begins with sensory stimulation, where raw data is analyzed hierarchically. Hubel and Wiesel’s work on feature detectors (simple, complex, and hypercomplex cells) illustrates how the visual system builds complex images from basic features. J. J. Gibson’s ecological approach (1966, 1979) supports this by arguing that perception is "direct," relying on invariant information in the environment (affordances) rather than internal inference.

Top-down processing, conversely, is driven by prior knowledge, expectations, and context. Richard Gregory (1970), extending Helmholtz’s idea of unconscious inference, described perception as hypothesis testing that makes sense of ambiguous stimuli. This is demonstrated by perceptual set, such as in Bruner and Minturn’s (1955) study where the same symbol was perceived as a "B" in a letter context but a "13" in a number context.

Critically, neither process is sufficient alone. Bottom-up processing cannot explain how degraded or ambiguous input, as in low visibility, is still recognised, while top-down processing can be "fooled" by illusions or strong expectations, leading to perceptual errors.

In everyday life, we utilize a combined approach. For instance, when reading a handwritten note, bottom-up processing identifies the ink strokes, while top-down processing uses linguistic context to identify the words. Similarly, in driving, we use bottom-up data to detect a red light, but top-down knowledge of traffic rules dictates the action of stopping. Treisman’s Feature Integration Theory (1980), in which attention binds bottom-up features, and McClelland and Rumelhart’s Interactive Activation Model (1981), in which word-level knowledge feeds back to letter detectors, formalize this synergy. The relative weighting of these processes shifts based on stimulus quality and expertise; for example, a radiologist relies more on top-down pattern recognition than a novice, who must rely on bottom-up feature analysis.

(b) Why psychology is more scientific than other social sciences Psychology is often regarded as more scientific than other social sciences such as sociology or political science because it relies on experimental control, precise measurement and biological anchoring. Where other social sciences depend largely on correlational, historical or observational data, psychology has prioritised the experiment to establish causality.

This status is rooted in its early adoption of quantitative methods. Fechner’s psychophysics (1860), building on Weber’s law, expressed the relation between physical stimulus and sensation mathematically. Wilhelm Wundt’s Leipzig laboratory (1879) marked the shift from philosophy to empirical science, and Ebbinghaus (1885) derived the forgetting curve through controlled self-experimentation.

Modern psychology manipulates an independent variable (IV) to observe its effect on a dependent variable (DV) while controlling extraneous variables through control groups and randomisation. Pavlov’s conditioning studies, Asch’s conformity experiments and Bandura’s Bobo doll study exemplify this. Psychometrics adds rigour: tests are standardised, normed and checked for reliability (consistency) and validity (accuracy). Neuroscience methods such as EEG and fMRI further anchor behaviour in measurable biology.

This status is not absolute. The Open Science Collaboration (2015) replicated only about one-third of 100 published studies; demand characteristics (Orne, 1962) and WEIRD samples (Henrich et al., 2010) limit validity and generalisability. Yet falsifiability (Popper), replication, meta-analysis and pre-registration keep psychology self-correcting, distinguishing it from more interpretive social disciplines.

(c) Sampling bias, data analysis and interpretation Sampling bias occurs when the sample used in a study is not representative of the target population, systematically distorting the accuracy of research outcomes. It is distinct from random sampling error, which decreases as sample size increases; bias, however, persists regardless of sample size. Common types include volunteer bias (participants self-selecting), undercoverage (missing segments of the population), and non-response bias. A classic example is the 1936 Literary Digest poll, which predicted the wrong presidential outcome because its sample was drawn from telephone and car owner lists, over-representing affluent voters. This bias primarily threatens external validity (generalizability). Remedies include using probability sampling designs—simple random, stratified, cluster, or systematic sampling—and ensuring an adequate sampling frame.

Data analysis and interpretation are the final steps in generating meaningful findings. Analysis involves statistical procedures to determine if results are significant, focusing on Type I (false positive) and Type II (false negative) errors, statistical power, and effect size. However, analysis alone is insufficient; interpretation must be rigorous to avoid fallacies.

Researchers must avoid the "correlation does not imply causation" error. They must also avoid the ecological fallacy (inferring individual traits from group data) and the confusion between statistical significance (a small *p*-value) and practical significance (a meaningful real-world effect). Meaningful findings are generated only when statistical data is interpreted within the context of the theoretical model and the specific limitations of the study.