2.2 Observational and Survey Techniques

Observational and Survey Techniques

Observational and survey techniques give social psychologists ways to study behavior and attitudes without manipulating variables. These methods are especially useful for exploring how people act in real-world settings or for measuring beliefs and opinions across large groups. Both approaches come with specific biases that can threaten data quality, so understanding how to design and implement them well is a core skill in research methods.

Observational Techniques

Types of Observational Methods

Naturalistic observation means watching people (or animals) in their everyday environment without interfering. The researcher stays unobtrusive so that the behavior being recorded is as authentic as possible. A classic social psychology example: observing how strangers interact in a crowded subway to study helping behavior. The strength here is ecological validity, meaning the findings reflect what actually happens in real life. The trade-off is that you have little control over what occurs, so it's harder to draw causal conclusions.

Participant observation takes things further. The researcher joins the group being studied and experiences events alongside them. This can be:

• Overt: the group knows the researcher's role
• Covert: the researcher's identity is concealed

Covert participation can yield more authentic data since people aren't performing for a researcher, but it raises serious ethical concerns about informed consent. Either way, participant observation provides rich, detailed insight into group dynamics that you can't get from the outside.

Structured interviews use a predetermined set of questions asked in the same order to every participant. This consistency makes it easier to compare responses across people. Questions can be open-ended ("Describe how you felt") or closed-ended ("On a scale of 1–5, how anxious were you?"). Interviewers need training to deliver questions the same way each time and avoid accidentally leading respondents toward certain answers.

Challenges in Observational Research

Reactivity is the broad problem of participants changing their behavior because they know they're being watched. The Hawthorne effect is the most well-known version of this: workers at the Hawthorne factory in the 1920s increased productivity simply because they were aware of being studied, not because of any actual change in conditions. Reactivity can make your data less accurate, which is why researchers try to be as unobtrusive as possible or allow an adjustment period before recording data.

Observer bias is a separate issue on the researcher's side. Observers may unconsciously notice or record behaviors that confirm their hypothesis while overlooking contradictory evidence. Two strategies help reduce this:

• Standardized coding systems: clearly defined categories for what counts as a specific behavior, so there's less room for subjective interpretation
• Multiple independent observers: having two or more people code the same behavior and then checking inter-rater reliability (the degree to which their ratings agree)

Survey Design

Questionnaire Development

Surveys let researchers collect data from large numbers of people relatively quickly. They can be administered in person, by mail, online, or by phone, each with different response rates and practical trade-offs.

Writing good survey questions is harder than it sounds. A few key principles:

• Avoid leading questions that push respondents toward a particular answer ("Don't you agree that...")
• Avoid double-barreled questions that ask about two things at once ("Do you find this course interesting and useful?" forces one answer for two separate judgments)
• Keep wording clear and concise so respondents interpret the question the way you intended
• Pilot test the questionnaire on a small group first to catch confusing or ambiguous items before the full study

The Likert scale is one of the most common response formats. It typically offers 5 or 7 points ranging from "Strongly Disagree" to "Strongly Agree." This format lets you capture degrees of opinion rather than forcing a simple yes/no, and the numerical values allow for statistical analysis. Researchers often combine multiple Likert items into a composite score to measure complex constructs like self-esteem or job satisfaction.

Survey Biases and Limitations

Social desirability bias is the tendency for people to answer in ways that make them look good rather than answering honestly. For example, respondents might underreport prejudiced attitudes or overreport how often they exercise. Strategies to reduce it include guaranteeing anonymity, wording questions indirectly, and including a social desirability scale (a set of items designed to flag respondents who are likely giving overly favorable self-reports).

Question order effects happen when earlier items influence how people respond to later ones. If a survey first asks about recent personal failures and then asks about overall life satisfaction, the priming from those failure questions can drag satisfaction ratings down. Randomizing question order across participants helps control for this.

Response sets are patterns in how people answer regardless of question content:

• Acquiescence bias: the tendency to agree with statements no matter what they say. Including a mix of positively and negatively worded items helps detect this.
• Extreme responding: consistently choosing the most extreme option on every scale, which can distort results.

Sampling Considerations

Sample Size and Statistical Power

Your sample size directly affects whether your study can detect real effects and whether your findings generalize to the broader population.

• Larger samples produce more precise estimates of population characteristics and reduce the impact of individual outliers.
• Power analysis is a calculation done before data collection to determine the minimum number of participants needed. It takes into account the expected effect size (how large the phenomenon is), the significance level (usually .05), and population variability.
• Margin of error shrinks as sample size grows. The formula is:

$ME = z \times \sqrt{\frac{p(1-p)}{n}}$

where $z$ is the z-score corresponding to your confidence level, $p$ is the estimated population proportion, and $n$ is the sample size. In practice, researchers balance the desire for precision against real constraints like time, budget, and participant availability.

Sampling Methods and Representativeness

How you select participants determines how confidently you can generalize your results. The two broad categories are probability sampling (every member of the population has a known chance of being selected) and non-probability sampling (no random selection, which introduces potential bias).

Probability sampling techniques:

• Simple random sampling: every individual has an equal chance of selection, like drawing names from a hat
• Stratified sampling: the population is divided into meaningful subgroups (e.g., by age or gender) and then random samples are drawn from each subgroup, ensuring all groups are represented
• Cluster sampling: entire groups (e.g., classrooms or neighborhoods) are randomly selected, and everyone within those groups participates. This is more practical when you can't easily list every individual in a population.

Non-probability sampling techniques:

• Convenience sampling: selecting whoever is easiest to reach (e.g., college students in an intro psych course). This is extremely common but limits generalizability.
• Snowball sampling: current participants recruit others from their networks. This is useful for hard-to-reach populations (e.g., undocumented immigrants) but can produce a sample that's too homogeneous.
• Quota sampling: the researcher sets targets for specific demographic categories and recruits until those quotas are filled. It resembles stratified sampling but without random selection.

Representativeness is the degree to which your sample mirrors the target population on key characteristics like age, gender, ethnicity, and socioeconomic status. When certain groups are over- or under-represented, selection bias creeps in, and your conclusions may not apply beyond the specific people you studied. This is why researchers always discuss sampling limitations when reporting results.