Midterm Content Flashcards

Question

What are some search tips when searching a database? [7]

Answer 1

* Use PICO framework * Search one concept at a time * Appropriate use of MeSH terms * Keyword strategies * Combining with AND & OR appropriately * Applying limits at the end of the search * Reverse engineering

Answer 2

* BG: general knowledge about a topic; can usually be answered by a 2º or 3º literature source (e.g., review article, clinical guidelines, textbook, etc. * FG: patient-specific; focused question; PICO is used to formulate a clinical question

Answer 3

Medline is the premier biomedical database with over 26 million journal articles in the life science. There are several different interfaces to the database (e.g., PubMed, Ovid) Searching a database involves both a subject heading (MeSH) strategy as well as a keyword strategy.

Answer 4

Medical subject heading Controlled vocabulary indexing the article's content; subject specialists index the article by tagging it with controlled vocabulary from a standardized list. Organized in a hierarchical structure which allows for searching at various levels of specificity. Articles are indexed to the most specific MeSH (e.g., articles about breakfast will not necessarily be indexed as well with meals)

Answer 5

* It takes a few months for articles to be indexed with MeSH * Sometimes an appropriate MeSH is not available * The concept is new to literature or is only recently added as a MeSH (e.g., Vegan / (2016)) * Indexes errors/omissions * Completeness of your search

Answer 6

1. **Synonyms** to describe concept 2. Join synonyms with **Boolean** operator: OR (e.g., fat OR lipid) 3. **Truncation**: \* (e.g., diet\* = diets, dietary, diet) 4. **Adjacency operator:** adj (e.g., restricted adj2 fat\* = the word restricted is within 2 words of the word fat)

Answer 7

* Bias or systematic (spurious) * Effect-cause (real) * Confounding (real) * Chance/random error (spurious) * Cause-effect (real)

Answer 8

**A confounder is a 3rd factor that is:** * associated with the exposure and the outcome, and * makes the two appear related when they may not be.

Answer 9

* _Study design_ * **Specification**: design _inclusion criteria_ that specify a value of a potential confounder and exclude those with that value * **Randomization**: aim to _evenly distribute confounders_ among study * _During analysis_ * **Stratification**: group and analayze participants based on their _level of confounding_ variable * **Statistical adjustment**: techniques used to _adjust_ for confounders using available software *Caveat: data on confounding factor must be collected during the study to do stratification analysis/statistical adjustment (importance of good study design).*

Answer 10

A 3rd variable _modifies_ (does not explain) the effect. The strength of the apparent association varies over different categories of a third variable (e.g., age, gender, genotype)

Answer 11

* **Confounder** → _explains_ association * **Effect modifier** → influences association, _does not explain_ it.

Answer 12

**Population**: entire group of interest (generalized results apply to this population) **Sample**: portion of population included in the study (findings from this sample should be generalizable to the population)

Answer 13

1. To choose a sample _representative_ of the population. 2. So results can be _generalized_ to the population 3. To reduce the _sampling error_.

Answer 14

Selected individuals over-/under-represent certain population attributes related to the study.

Answer 15

**Sampling error = x̄ - μ** More of a theoretical concept since we can't know the true mean of the population. However, it can be estimated. Sampling error is the difference between the sample and population means. Reducing this error is the main purpose of sampling. As the sample size increases, sampling error generally decreases (i.e., larger sample sizes are more representative of the population)

Answer 16

**Too small** → significant differences may not be found; sample is not representative of population. **Too large** → very expensive

Answer 17

**Quantitative sample size** → must be large enough to draw inferences from; the larger, the more representative (unless using poor sampling technique); note, a small sample the is representative is better than a large sample that is unrepresentative. **Qualitative sample size** → small numbers provide in-depth information; e.g., proceed until saturation (continue until you're not hearing anything new)

Answer 18

1. Define the target population (i.e., who you want the findings to generalize to); define the sampling frame (i.e., accessible population from which sampling is drawn); set inclusion/exclusion criteria 2. Recruit/select subjects via probability or non-probability sampling.

Answer 19

Probability → random selection; reduces sampling bias; preferred Non-probability → non-random selection

Answer 20

**Probability** → Simple random, stratified random, cluster, systematic **Non-probability** → convenience, purposive, quota, snowball NOTE: No matter the method, the final sample = recruited subjects who _consent_ to participate.

Answer 21

A simple random sample is a type of probability sampling where every individual within the sampling frame has an equal and independent chance of selection. 1. Define population. 2. List everyone in it. 3. Choose randomly.

Answer 22

A stratified random sample is a probability sampling method used if people in the population differ systematically along some characteristic relating to the study factor. It may be disproportionate or proportionate. 1. Sampling frame is divided into strata (e.g., age, sex, ethnicity) 2. Samples are drawn randomly from each strata.

Answer 23

A type of probability sampling where researchers choose every k^th name on a list (more simple than simple random sampling); will produce a random sample unless the subjects are listed in an order that could affect outcomes. 1. Divide population by the size of the desired sample (e.g., 50 (population)/10(sample size) = 5) 2. Select a random starting point. 3. Select every 5th name as a sampling interval.

Answer 24

**Disproportionate** → population proportions are not preserved; higher proportions selected from some groups compared to others; useful when it may be difficult to reach a sufficient sample size with some groups; data must be weighted during analysis. **Proportionate →** population proportions are preserved 1. Characteristics of interest are identified (e.g., gender) 2. Individuals in population are listed separately according to classification 3. Proportional representation of each classification is determined (or not) 4. A random sample is selected that reflects the determined proportions OR randomly choose the same number from each group regardless of population proportions.

Answer 25

Cluster sampling is a type of probability sampling used if natural groupings exist in the population (e.g., school districts, geographic boundaries). It divides the population into feasible samples. Instead of randomly sampling individuals, entire units/groups are identified and a random subset (cluster) of them is selected, and then finally individuals from each cluster are chosen at random. e.g., Of all schools in BC → 10 selected randomly → of selected schools → 10 students collected randomly

Answer 26

This is a non-random sampling technique is when researchers recruit whoever they can, or the most easily sampled population. This type of sampling has very weak representativeness, and may show bias due to ‘self-section’. For example, standing at a mall or a grocery store and asking people to answer questions would be an example of a convenience sample. Or posting an advertisement and recruiting any respondent.

Answer 27

This is a type of non-probability sampling where researchers select subjects specifically for a reason → for example, if they know they need a very specific type of person, they will be sought out. The sample will be selected by the researcher based on _specific (subjective)_ criteria → more often used in qualitative research (rare in quantitative)

Answer 28

This type of non-probability sampling is ‘equivalent’ to stratified random sampling. It is used when stratified random sampling is not possible. Participants with the characteristic of interest are non-randomly selected until a quota is met.

Answer 29

This is a type of non-probability sampling where existing subjects recruit future subjects, which may be useful for studying ‘rare’ or hard-to-reach groups. This type of sampling may be biased by the social networks of the original subjects.

Answer 30

**Convenience** → whoever is available based on objective inclusion/exclusion criteria (e.g., poster ad for students age 18 - 25) **Purposive** → sample selected by researcher based on some subjective criteria; sample is at discretion of researcher (i.e., ‘hand-picked’ selection of _info-rich sources_); used almost exclusively in qualitative research

Answer 31

**Conceptualization** → the process of specifying what we mean by a term; ‘concepts to observations’; helps translate the abstract theory/construct into specific variable(s) (i.e., operationalize). **Operationalize** → the process of connecting concepts (i.e., what we want to observe) to observations (i.e., measurable). Operational definitions: How will the variable(s) be measured? What level of measurement will be used?

Answer 32

1. **Nominal** → categories with no ranks 2. **Ordinal** → ranked order 3. **Interval** → ranks with known intervals between them 4. **Ratio** → has an absolute zero and meaningful ratios *(think: NOIR)*

Answer 33

Assignment of labels; used for categorical variables. The categories vary in quality but not amount (i.e., CANNOT say one is more/less than another) e.g., Did you walk/cycle/transit/drive to UBC on Friday?

Answer 34

Assignment of values along some underlying dimension; one observation is ranked above/below another; variables are ordered with no defined interval (i.e., CANNOT say the amount of one variable is more/less than another) e.g., underweight/health weight/overweight

Answer 35

Assignment of values with equal distances between points; one score differs from another on some measure that has equally appearing intervals. The zero is ARBITRARY, so no meaningful ratios can be used (e.g., 1980-81 has the same interval as 2010-11 BUT 0 does not imply the beginning of time). e.g., temperature (i.e., 0°C does not mean absence of temperature)

Answer 36

Values have ranks and equal distances between points (i.e., meaningful ratios can be computed); CAN say that values differ, by how much, and what this means due to an absolute zero point. This is the highest level of measurement. e.g., age, weight, height, caloric intake

Answer 37

**Nominal** → sourced from the sun or from the diet **Ordinal** → does not meet / meets/ exceeds DRI **Interval** → set RDA as 0 and report scores as above or below (e.g., RDA 20, intake 15 → report -5) **Ratio** → Vitamin D intake reported in mcg/day

Answer 38

* How a variable is measured can determine the amount of information obtained (e.g., BMI vs. under/normal/overweight). * How a variable also determines the type of statistical tests that can be used.

Answer 39

1. The _level of original measurement_ matters. 2. There may be a _minimal detectable difference_ (i.e., can only detect differences greater than the measurement error of the instrument used) 3. _Starting points_ matter → floor/ceiling (i.e., already as high/low as it can get, outcome won't differ) effects; regression to the mean) 4. _Variables change naturally_ over time (e.g., blood pressure) 5. _Tools of measurement_ must be both _reliable_ and _valid_.

Answer 40

**Reliability** → tool is consistent **Validity** → tool measures ‘what it should’ These are important for knowing if we are measuring what we want to measure, and if we are doing so correctly.

Answer 41

Reproducibility of a measurement (i.e., the extent to which a measurement is _consistent_ and _free from error_).

Answer 42

**Systematic** → ‘predictable’ errors of measurement (e.g., consistent under-/over-estimation); major concern for VALIDITY of measure. **Random** → due to chance; major concern for RELIABILITY of measure (i.e., there is no consistency)

Answer 43

1. **Tester or rater** → errors in taking measurements, recording behaviour or data entry 2. **Measurement instrument** → equipment malfunction; uncalibrated equipment; unclear questionnaire 3. **Variability in characteristic being measured** → transient states of participants (e.g., mood, health, fatigue-level, blood pressure) 4. **Situational factors** → room temperature, lighting, crowing, etc.

Answer 44

1. Maintain consistent scoring procedures; ensure testers are trained uniformly 2. Increase the number of items/observations; standardize instructions; eliminate unclear questions 3. Standardize testing conditions 4. Minimize the effects of external events

Answer 45

1. Inter-/intra-rater 2. Test-retest 3. Parallel forms 4. Internal consistency

Answer 46

**Inter-rater** → have two raters judge the same event/behaviour and assess agreement **Intra-rater** → assess the stability of measures by the same person

Answer 47

Give the test to the same people at two time points and see if the results are the same. It is necessary to be aware of _testing-effects_ (i.e., change just by being tested), and be sure that measured characteristic does not change over time.

Answer 48

Give two different but equivalent forms of the test to the same group and compare the results. e.g., survey by paper vs. online; survey in 1st and 2nd language of participant

Answer 49

The correlation coefficient, which indicates how scores on one test change relative to scores on a second test. +1 = perfect reliability | 0 = no reliability | -1 = inverse Also the intraclass correlation coefficient may be used. The results from the two tests must be related (correlated) and agree.

Answer 50

Extent to which items in the tool measure the same characteristic (and nothing else); degree of consistency/correlation among items. Internal consistency may be measured by: 1. **Split-half reliability** → randomly divide items into 2 subsets and examine the consistency in total scores (e.g., results from 100 question questionnaire must agree when split into Q1-50 and Q51-100) 2. **Cronbach's Alpha** → conceptually, the _average consistency_ across all possible split-half reliabilities; 0.7 is considered acceptable

Answer 51

1. Face 2. Content 3. Criterion 4. Construct *(think: FCCC)*

Answer 52

‘face-value’ Does the measuring instrument appear to test what it is supposed to? Does it appear to be valid to those completing it? Answers may not be honest/accurate if not.

Answer 53

How well do the items represent all relevant items? Ask an expert: Does this instrument measure everything it's supposed to? Is anything missing? Is my question comprehensive?

Answer 54

Ability of the tool to predict results obtained on a valid external criterion or reference standard; criterion/standard should be a valid indicator of variable of interest. Two types: 1. **Concurrent (present)** → e.g., FFQ validated with 24 hour recalls or diet records collected for the same time period 2. **Predictive (future)** → e.g., using HbA1C to predict diabetes risk

Answer 55

Construct validity cannot be directly measured due to abstract concepts like healthy eating and happiness. The extent to which test results are related to underlying abstract concept (difficult to establish; often used in combination with other validity evaluation methods). _Strategies [2]:_ * Show that people with and without certain traits score differently (i.e., they can be differentiated). * DISCRIMINANT → two unrelated measures should not be correlated * Compare the measure with other related measures of similar or differing constructs. * CONVERGENT → two measures of a similar construct should be correlated

Answer 56

**Reliability** → ensure experiments can be repeated and give the same results; often run duplicate or triplicate analysis **Validity** → positive control (a known amount of something you are trying to measure or detect is run with assay to ensure you can detect it and measure it accurately); negative control (ensures you only detect what you want and not something else)

Answer 57

**Parameter** → measure of a population characteristic

Answer 58

Statistic → measure of a sample characteristic

Answer 59

Summarizes the data and provides basic information about the distribution of scores in a sample using the mean, min, max, range, and standard deviation.

Answer 60

Inferences from the sample to a larger population.

Answer 61

Treatments or conditions under control of the researcher; the researcher may not have ‘direct’ control, but might assign groups based on this feature.

Answer 62

Outcomes of the research study; sensitive to changes in the independent variable

Answer 63

Numerical variables have measurable quantity; two types: 1. **Discrete** → variable can take on a finite number of values (e.g., # of students in FNH398) 2. **Continuous** → values can range along a continuum (e.g., height) Level of measurement: ratio or interval

Answer 64

Categories/groups; measures of proportions (e.g., gender, ethnicity) Level of measurement: ordinal, nominal

Answer 65

Includes figures/graphs, measures of central tendency and variability as pivotal first steps for visualizing and understanding data and their distribution. **Central tendency** → mean (normal distributions), median (skewed distributions), mode (categorical variables) **Spread/variability** → range (crude), standard deviation (normal distributions), percentiles and inter-quartile ranges (skewed distributions)

Answer 66

* Symmetrical * Asymptotic tail * Bell shaped histogram

Answer 67

**Mean** → arithmetic average; normal distributions **Median** → midpoint of distribution; skewed distributions **Mode** → most frequently occurring score; categorical data

Answer 68

**Normal** → mean = median **Positive skew** → long right-hand tail; mean \> median **Negative skew** → long left-hand tail; mean \< median

Answer 69

The degree of spread/dispersion in a set of scores; central tendency doesn't tell us anything about spread (i.e., two curves with the same mean may have differing degrees of variability)

Answer 70

1. **Range** = max - min; crude indication of spread; does not delineate shape of distribution or how much scores vary from the mean 2. **Standard deviation** = average difference of scores from the mean; sensitive to outliers; not the same as standard error 3. **Standard error** = uncertainty in measurement of the mean; portrays uncertainty around the mean; gives a sense of where the true population mean lies; measure of variation of the mean 4. **Percentiles** = value below which a certain percent of the data lie (e.g., 25^th percentile = 25% of the data lies below this number) 5. **Quartiles** = divide the data into four equal parts; inter-quartile range = difference between 3^rd and 1^st quartiles; Q2 = median; Q1 = 25^th percentile; Q2 = 50^th percentile; Q3 = 75^th percentile

Answer 71

The # of standard deviations a score is away from the mean.

Answer 72

**99.7%** of the data lies between -3 and 3 (i.e., 99.7% of the data lies within 3 standard deviations from the mean) **95%** of the data lies between -2 and 2 **68%** of the data lies between -1 and 1 The following is also mirrored on the positive half of the distribution: 0 to -1 → 34% - 1 to -2 → 13.5% - 2 to -3 → 2.35% - 3 to -∞ → 0.15%

Answer 73

An interval in which the true population is likely to exist. e.g., For a 95% confidence interval implies that there is a 95% probability that the true population mean exists within this interval.

Answer 74

If it lies above the 95^th percentile or below the 5^th percentile.

Answer 75

When we find a difference between group means in our sample we want to know: Does this represent a real difference between target groups? Is this likely just a chance/random finding in our sample? Inferential statistics is used to determine the likelihood of the finding happening by chance.

Answer 76

Null → states that there is _no difference_ between groups (i.e., H_o: μ₁ = μ₂); always stated using population parameters; accepted as true in the absence of other information. Alternative → states that there _is a difference_ between groups or relationship between variables (i.e., H_A: μ₁ ≠ μ₂)

Answer 77

Statistical tests to either: Reject the null (statistically significant difference between groups) or, Fail to reject the null (no statistical significance between groups)

Answer 78

‘False positive’ The null hypothesis is rejected BUT there is no real difference between groups. **α = probability of making a Type I error (set to 0.05 usually; 5% chance we reject the null when it is actually true**

Answer 79

‘false negative’ The null hypothesis is NOT rejected but there IS a real difference between groups or association between variables. β = probability of making a Type II error; usually set at 0.2 (20% chance of failing to reject the null when it is false)

Answer 80

The ability of a test to reject the null hypothesis (i.e., the likelihood of finding a difference _if one exists_) **power = 1 -** **β** For β = 0.2, power = 0.8; 80% power to detect a difference if one exists.

Answer 81

1. State the null and alternate hypothesis. 2. Establish a significance level (i.e., α) 3. Choose the most appropriate test. 4. Collect the data. 5. Analyze data using statistical test. 6. Reject or fail to reject the the null

Answer 82

Independent → dependent → statistical test Categorical → Continuous → t-test (2 categories) or ANOVA (\> 2 categories) Categorical → Categorical → Chi-squared Continuous → Continuous → Correlation

Answer 83

Large differences between means are [**less likely**] to occur by chance alone than small differences between means. The greater the variability in distributions, the [**more likely**] the difference between the means is due to chance.

Answer 84

1. Mean for each group 2. Standard deviation for each group 3. Sample size

Answer 85

Probability of obtaining our result (by chance) if the null is assumed to be true; calculated by probability distribution

Answer 86

T-test → t-value compared to t-distribution ANOVA → f-value compared to f-distribution

Answer 87

Based on α (i.e., level of significance) **P-value \> α** (fail to reject; probability that results are due to chance is greater than we are willing to accept) **P-value \< α** (reject; unlikely results are due to chance alone; likely represents a real difference; statistical significance)

Answer 88

**Parametric** → representative of population; data approximately normally distributed; variances of groups being compared are approximately equal; interval or ratio scales of measurement **Non-parametric** → assumptions of parametric tests are not met; any distribution including skewed

Answer 89

**T-test** → 2 group means **ANOVA** → \> 2 group means; only tells us if significant differences occur anywhere among groups, but not which specific groups differ → Post-hoc analysis used to see ‘which’ groups differ (i.e., Fischer's least significance test, Tukey's Honestly significant difference test)

Answer 90

1. **Two-tailed** → used for non-directional hypotheses; mean of one group may be larger or smaller than the mean of the second group 2. **One-tailed** → stringent; harder to find statistical significance; less Type I error chance; used when direction of change is known (e.g., mean 1 \> mean 2); used for directional hypotheses 3. **Independent means** → used when two independent, unrelated groups are compared 4. **Dependent means** → a.k.a. paired t-test; used when groups are related (e.g., difference in F+V intake on Monday and Saturday → same individuals both times)

Answer 91

more comparisons = greater chance of error in t-test

Answer 92

1. **One-way** → one independent variable (e.g., measuring BP across groups of sodium intake) 2. **Two-way** → two independent variables (e.g., measuring BP across groups of sodium intake and also groups of magnesium intake) 3. **Repeated measures** → e.g., measuring blood pressure in the same person after three doses of sodium

Answer 93

x-y scatterplot; correlation coefficient (for linear relationships) Pearson correlation varies between -1 and 1. Direction of relationship indicated by +/-. Strength of relationship indicated by value. If X… → If Y… → The correlation is… ⇡ → ⇡ → + (direct) ⇣ → ⇣ → + (direct) ⇡ → ⇣ → - (inverse) ⇣ → ⇡ → - (inverse)

Answer 94

0. 8 - 1.0 → very strong correlation 0. 6 - 0.8 → strong correlation 0. 4 - 0.6 → moderate correlation 0. 2 - 0.4 → weak correlation 0. 0 - 0.2 → no/very weak correlation

Answer 95

1. Cannot extrapolate outside the range of the data. 2. Need a range of values to show a correlation. 3. Correlation coefficient does not tell us if relationship is statistically significant (additional tests needed to determine P-value)

Answer 96

**Correlation** → direction (+/-) and strength (strong/weak) **Regression** → magnitude of change; how much does y vary when x varies?

Answer 97

Line of ‘best fit’ (y = mx + b) Can be used to predict a value for the dependent variable (y) for a given value of the independent variable (x).

Answer 98

R² The proportion of variability in a data set that is accounted for by the statistical model → ranges from 0 to 1. If 0 → none of the variation in the outcome is explained; x has no impact on y If 1 → the regression line ‘perfectly fits’ the data; all variation in outcome is explained Example → R² = 0.8 → 80% of the variation in the dependent variable (y) is explained by the independent variable (x).

Answer 99

Used to describe the association between several independent variables simultaneously, and a single dependent variable; predictor variables can be categorical (using dummy variables like 0 = yes, 1 = no) or continuous

Answer 100

Dependent/outcome variable is **dichotomous** (yes/no). Independent variable may be continuous or categorical.

Answer 101

Independent/predictor variable → continuous OR categorical Dependent/outcome variable → continuous

Answer 102

Independent/predictor variable → continuous OR categorical Dependent/outcome variable → categorical

Answer 103

1. Independent (unpaired) t-test 2. Dependent (paired) t-test 3. One-way/Two-way ANOVA 4. Repeated measures ANOVA 5. Pearson correlation

Answer 104

1. Mann-Whitney U-test 2. Wilcoxon signed ranks 3. Kruskal-Wallis 4. Friedman 5. Spearman rho

Answer 105

Used when all variables are categorical. Compares the frequency with which we would _expect_ certain observations to occur with the frequency that it _actually occurred_.

Answer 106

Considers multiple variables; may be used to ‘adjust’ or ‘control’ for the influence of several factors.

Answer 107

**Statistical significance** → the probability that the observed result occurred by chance is low. **Meaning** → clinical significance; requires good judgement to determine whether results represent a meaningful difference.

Midterm Content Flashcards

(133 cards)