Internal and External Validation

Internal validity threats are procedures, treatments or experience of participants that threatens the researcher’s ability to draw correct conclusion (Creswell, 2014). Some of the internal validity threats are:

1. Participants of research – history (events impacting while conducting research), maturation (research participants may mature or change), regression (participants with extreme behavior traits), selection (participants chosen with certain views), and mortality (participant may left the study);

2. Experimental treatment – diffusion (communication between participants during research may impact outcome), compensatory or resentful demoralization (participants may act differently due to monetary issues), and compensatory rivalry (monetary difference between control and experimental group may influence their views);

3. Experimental procedures – testing (participants may perform intentional actions based on past experiences lead to bias) and instruments (change or modification of instrument)

External validity threats result in incorrect interference from experiment (Creswell, 2014). Some of the external threats are:

1. Interaction of selection and treatment – narrow characteristics of participants may lead to generalization issues to other individuals;

2. Interaction of setting and treatment – characteristics of experiment settings may influence outcome;

3. Interaction of history and treatment – experiment outcome being time-bound thus may not be generalized

The possible research idea is to mine the tweets and find insights on health trends, specifically seasonal outbreaks, like influenza in the given neighborhood. In this particular topic, some of the possible internal and external validity threats and procedures to minimize their impacts are:

1. Participant – selection of mature participants is desired to minimize false reporting of symptoms by using age filter;

2. Experimental procedure – pre-processing of social media data is needed to avoid data with spelling mistakes as well as some context filtering is needed which will ensure that tweets where person experiencing flu symptoms are analyzed instead of informational tweets;

3. Natural Language Processing (NLP) Algorithm – several NLP algorithms are available with varying results which is why one best algorithm need to be used throughout to avoid variance in outcome;

4. Statistical Validation – standard statistical measures are required to properly validate the performance of flu detection from social media data versus government authority like, Center of Disease Control (CDC)

Reference

Creswell, J. W. (2014). Research design: Qualitative, quantitative, and mixed methods approaches. Thousand Oaks, CA: Sage.