Where to go from here HOW TO FRAME YOUR RESEARCH Before diving into analysis, think carefully about what your data represents and what valid conclusions you can draw from it. This section Determining what conclusions you can draw helps establish the representativeness of a The process of learning and drawing conclusions dataset and pinpoint uncertainties to avoid from data is called “inference.” There are two inappropriate or incorrect conclusions. types of inference: Establishing what type of sample your data is There are three types of samples. The type of sample you have depends on how many units of analysis you have and how they were selected. ›› Complete sample: an entire population. Other terms for a complete sample are a “census” or a “complete enumeration”. ›› Probability sample: units in the population were selected through a probabilistic mechanism. This mechanism ensures that the probability of selection for every unit in the population is known. Convenience sample: units were selected by a non-probabilistic mechanism. For each unit in the population, the probability of being selected into the sample is unknown, while it is very high for some units and very low or zero for other units. In human rights research, convenience samples are the most likely type of available data samples. ›› Statistical inference draws conclusions about the population, based on the sample. These conclusions are valid only if you have a complete or probability sample. ›› Description draws conclusions about the sample you have. Making inferences from complete samples Complete samples are special in that they are identical to the population, so whatever you learn about the sample, you also learn about the population. Making inferences from probability samples Probability samples are representative of the population if the underlying sample probabilities are accounted for. Because each unit in the population has a known probability of being selected into the sample, on average, we can approximate the same values and distributions in our sample that exist in the population (minus a probability sampling error). We do not consider all units in the population, but only a representative sample thereof, so we call these conclusions “estimates”. 74