Sampling

Question 1

Epidemiological approaches require us to gather information about ___________.

Answer 1

populations

(health of a group; can’t test everyone —> sample)

Question 2

What are the main 3 approaches to epidemiological studies?

Answer 2

1. establish level of occurrence of a disease - prevalence
2. test hypotheses - changes in levels of occurrence in different populations, identification of risk factors
3. detection of disease - declare freedom from disease for trade

Question 3

What is the difference between a census and a sample?

Answer 3

CENSUS = every individual in the population is sampled giving an exact measurement of what is going on without the need for statistics ($$$)

SAMPLE = group from the population is used to make inferences about the population based on information from the sample (quick, $, practical)

Question 4

What is a target population? Source population? Sample?

Answer 4

TARGET POP = population you want to make inferences about with respect to your objectives (feral cats in NA)

SOURCE POP = subset of the target population from which you will draw your sample (feral cats on LI)

SAMPLE = individuals actually measured or enrolled in the study (feral cats spayed or neutered by LIU’s team in 2021)

Question 5

What is the difference between internal validity and external validity?

Answer 5

INTERNAL = whether or not the study results obtained from the study sample are valid for the source population

EXTERNAL = how well the study results can be generalized to the target population; subjective assessment of whether or not the source population and sample are broadly representative of the target population (are the results from the US relevant to a reader in Australia?)

Question 6

What is a sampling frame? Sampling unit?

Answer 6

SAMPLING FRAME = complete list of all sampling units required for most probabilistic sampling methods

SAMPLING UNIT = unit of interest (animals, litters, pens, farms)

Question 7

What are the 2 approaches to sampling?

Answer 7

1. probability sampling
2. non-probability sampling

Question 8

What is probability sampling?

Answer 8

formal process of random selection used to ensure representativeness of the sample where every element in the population has a known probability of being included in the sample

• statistical inference can be made about the population from which you drew your sample

Question 9

What are the 5 types of probability sampling?

Answer 9

1. simple random
2. systemic random
3. stratified
4. cluster
5. multistage

Question 10

What is the main characteristic of probability sampling?

Answer 10

every element in the population has a known non-zero probability of being included in the sample

Question 11

What is simple random sampling (SRS)? What does it require?

Answer 11

basis of most probability sampling methods where every sample has an equal probability of being included in the sample

list of all subjects (sampling frame)

Question 12

What are the 3 steps to the procedure of simple random sampling (SRS)?

Answer 12

1. generate a list of animals
2. assign a unique number (ID) to each animal
3. generate a list of random numbers assigned to the IDs using Excel or random number generators

Question 13

What are the main 2 limitations of simple random sampling (SRS)?

Answer 13

1. difficult to obtain a complete sampling frame with large groups of animals and actually collect them
2. costs may be higher than other methods

Question 14

How does systemic random sampling compare to simple random sampling?

Answer 14

• does not require a sampling frame, but need all animals present sequentially (processing, appointment book, file)
• do not need to know the number of sampling units, but need a rough guess

Question 15

What are the 3 steps to the procedure of systemic random sampling?

Answer 15

1. select the sampling interval, j
2. randomly select a number within the sampling interval (7)
3. sample every jth animal after that (7, 27, 47, …)

Question 16

What is the main limitation to systemic random sampling?

Answer 16

could be subject to bias in the interval

• if there are 5 pens and the producer takes one cow at a time from each pen in a sequence for them to run through a chute, you shouldn’t sample every 5th because only cows from one pen will be tested

Question 17

What is stratified random sampling?

Answer 17

a sampling frame is divided into groups, or strata, of defined common characteristics (breed, sex, age) and randomized within each stratum

Question 18

What ensures each strata in stratified random sampling are present and properly represented?

Answer 18

having a proportional total within the stratum to the total number of the sample

Question 19

In what situation should stratified random sampling occur? Why would it be better than simple random sampling?

Answer 19

if it is suspected a certain breed (Siamese) have a higher prevalence of disease over other breeds (DLH/DSH), so a fair representation of breeds in the study is needed to prevent bias in an estimate of prevalence

• population of 1000 cats, ~70 Siamese
• sample size of 100 cats, ~7 Siamese in study
• SRS = 10% chance of having 4 or less Siamese and 10% chance of having 10 or more Siamese

Question 20

What are 2 limitations to stratified random sampling?

Answer 20

1. need to know about stratification variable in advance (risk factor)
2. less efficient than SRS if the strata do not explain variance within the population (strata are not risk factors)

Question 21

What are the 2 most common sampling methods used in veterinary medicine?

Answer 21

1. cluster
2. multistage

Question 22

What is cluster sampling?

Answer 22

sampling frame is divided into logical aggregations (clusters) and a random selection of clusters is performed

• clusters are common in vet med: any group that shares common characteristics, like farms, pens, and litters

(pick a shelter at random and sample all the animals there)

Question 23

How do clusters compare to others in the population?

Answer 23

animals within a cluster are more similar than animals between clusters

Question 24

What are the 3 advantages to cluster sampling?

Answer 24

1. only require a sampling frame for groups, but not for the members within the group
2. fewer sampling units (clusters) are required to achieve sample size of study units
3. more economical than SRS

Question 25

What are the 2 limitations to cluster sampling?

Answer 25

1. higher variability that SRS, so larger overall sample size (units of concern) is required to achieve the same precision - animals within the same shelter tend to be similar; low hygiene = low overall health of all animals that's not indicative of the target population 2. when between-cluster variation is large relative to within-cluster variation = need to sample many more clusters

Question 26

What is multistage sampling? Why do we do this?

Answer 26

same as cluster sampling, but take a sub-sample of subjects within clusters animals within a cluster are more alike than animals between clusters, so it's more efficient to sample fewer animals

Question 27

How are multistage sampling clusters picked?

Answer 27

cluster is randomly selected and individuals within each cluster are randomly selected (primary sampling units) farm ---> sows ---> litters ---> piglets

Question 28

What are the 3 advantages to multistage clusters?

Answer 28

1. only require a list of clusters (farms, not animals; sites, not pens or fish) 2. most practical for large studies 3. can adjust the number of clusters and primary sampling units (PSUs) based on the cost of acquiring samples and running those samples

Question 29

What happens when within cluster variation is high and between cluster variation is high in multistage sampling?

Answer 29

high WITHIN CLUSTER variation = smaller number of clusters, with more PSUs = weak farm effect - need fewer farms, but more animals high BETWEEN CLUSTER variation = larger number of clusters, with fewer PSUs = strong farm effect - need more farms, but fewer animals

Question 30

What are the 2 main limitations of multistage sampling?

Answer 30

1. larger sample sizes (clusters) may be required to achieve the same precision as SRS due to dependence (similarity) between clusters 2. greater complexity of statistical methods to analyze and estimate sample size

Question 31

What are the 4 types of non-probability sampling?

Answer 31

1. judgement (haphazard) 2. convenient 3. purposive 4. targeted (risk-based)

Question 32

What is judgement sampling?

Answer 32

investigator chooses subjects that "represent" the population in the judgement of the investigator

Question 33

What is convenient sampling? When is it most appropriate?

Answer 33

investigator uses samples that are easy to obtain if it is not needed for a group to be representative of the source population

Question 34

What is purposive sampling?

Answer 34

investigator chooses samples based on attributes (known exposure, cases) and if random samples are taken from the group, then it is a probability sample from the source population

Question 35

What is targeted (risk-based) sampling?

Answer 35

specifically used in animal disease surveillance programs using a biased sample to help find rare diseases or determine freedom from disease (DOESN'T represent target population

Question 36

What are the 2 advantages to non-probability sampling?

Answer 36

1. convenient and inexpensive 2. appropriate under certain circumstances, like exploratory/preliminary investigation, homogenous populations, and sampling to determine if disease is present in a population

Question 37

What are the 2 disadvantages to non-probability sampling?

Answer 37

1. not possible to know the probability of each unit being selected (not appropriate for descriptive studies at the population level) 2. can lead to biased population estimates with no way to quantify the extent of the bias

Question 38

What are the 6 steps to the sampling decision process?

Answer 38

1. define question 2. define source population 3. develop sampling frame 4. specify sampling method and determine sample size 5. select samples 6. analyze results

Question 39

What are the 4 types of data types? What are some examples of each?

Answer 39

1. continuous - any real numbers (including decimals); temperature, weight 2. dichotomous - yes/no, 1/0; pregnancy, gender 3. categorical (nominal) - ordinal with more than2 possibilities; breed, cancer class 4. count - integers, all whole numbers

Question 40

What are the 2 types of graphs used in veterinary epidemiology?

Answer 40

1. histograms 2. boxplots

Question 41

What is the difference between binomial and normal distribution?

Answer 41

binomial distribution is discrete, whereas the normal distribution is continuous, meaning that the binomial distribution has a finite amount of events, whereas the normal distribution has an infinite number of events

Question 42

What is skewed distribution?

Answer 42

a distribution that is neither symmetric nor normal because the data values trail off more sharply on one side than on the other

Question 43

What are the mean, median, and mode?

Answer 43

MEAN = average of a data set MEDIAN = middle value when a data set is ordered from least to greatest MODE = number that occurs most often in a data set

Question 44

What is the standard error of the mean?

Answer 44

indicates how different the population mean is likely to be from a sample mean, telling how much the sample mean would vary if you were to repeat a study using new samples from within a single population

Question 45

What are confidence intervals?

Answer 45

the probability that a population parameter will fall between a set of values for a certain proportion of times

Question 46

How much of the data should fall within 2 standard deviations in the area under the normal curve?

Answer 46

95.5%

Question 47

What are the 2 types of statistical errors?

Answer 47

TYPE I (α) - effect present in the study, effect absent in nature TYPE II (β) - effect absent in study, effect present in nature (lack of power)

Question 48

What is statistical power?

Answer 48

probability of a study to find a true difference (1-β)

Question 49

What are the 4 possibilities if a study truly finds no effect?

Answer 49

(type II (β) error) 1. truly no effect 2. study design issues (wrong dose, route of infection) 3. sample size too small (power < 0.80) 4. bad luck

Question 50

What is a p-value? When are we reasonably sure that the effect detected is not due to chance?

Answer 50

(type I (α) error) probability that a difference (as large as the one observed) could be due to chance P < 0.05 (we are 95% certain effect detected is a true effect)

Question 51

What is the standard error? What is it related to?

Answer 51

measure of the precision of the point of estimate variance, standard deviation, sample size

Question 52

What is a significance (hypothesis) test?

Answer 52

determines if the point estimate is significantly different from some value specified by the null hypothesis test (p-value)

Question 53

What is a confidence interval? When does it suggest a non-significant effect?

Answer 53

range of likely value for the point of estimate (95% CI) - range where the estimate falls 95% of the time if 95% CI includes the null value (picture = significant effect)

Question 54

EXAMPLE: visiting a herd of 100 cows, producer wonders if there is more mastitis on their farm than normal (the normal prevalence is 10%)

Answer 54

Question 55

EXAMPLE: visiting a herd of 100 cows, produce says they are producing an average of 30 kg/day (normal milk production is estimated 28 kg/day)

Answer 55

Question 56

What 3 things are being done with sample size calculation?

Answer 56

1. establish level of occurrence of a disease by estimating a proportion or mean with desired precision (level of FeLV in feral cats in LI) 2. test hypotheses by comparing 2 proportions or means (cats of placebo vs. those on treatment) 3. detection of disease/declare freedom from disease

Question 57

It's impossible to test all the animals in a population to be certain there is no disease. So, what is done?

Answer 57

test to a theoretical expected prevalence (level) of disease - "design prevalence" - if a disease was present in the population, you'd expect 1% of the animals to be infected, so you need to sample enough to be reasonably confident that you would detect the disease if the prevalence was 1% or higher