Quantitative Methodology - Science topic

You could look at the Mixed Methods Appraisal Tool.

Hong, Q. et al. (2019). Improving the content validity of the mixed methods appraisal tool: a modified e-Delphi study. Journal of clinical epidemiology, 111, 49-59.

Hi, I am facing the same issue with my data. How did you resolve it?

All due respect to you for this question

Thanks for having me. Quantitative is carrying numerical and statically way of various modular approach exercise. Subject, what nature and qualification of candidates you would like to take on board for modular education or information.

If you'd like to seek further discussion you can contact me on Whats app 00447954658263

Cronbach's alpha are different between factors dep and indep

teste de reliability

Hi Ivan Beroš - is the nature of the questions very sensitive? I ask because generally you can offer to anonymise the results when you code the results and store any results with identifiers in an institution's secure cloud server.

Very best wishes, James.

a) LPA could be appropriate if you assume that there are qualitatively distinct profiles of means across the indicators of Construct A across previously unknown subgroups (latent classes). That is, the mean profiles to be extracted should not just differ in level (in which case a dimensional [e.g., factor] model would probably be more useful), but also in shape (i.e., at least some of the mean profiles should cross in indicating group [latent class] differences in kind rather than just differences in degree). You could then examine whether class sizes vary across manifest grouping variables (e.g., group sexual violence vs. no violence) using multigroup LPA or whether class membership depends on continuous covariates (e.g., age) using covariate analysis {logistic regression of class membership om covariates).

b) Yes, Mplus does allow you to run LPA.

To answer this valuable question, let me repeat what I have written in my recording sheet during my bachelor study:

"Controlling" for age does not solve all problems. For example, see https://meehl.umn.edu/sites/meehl.umn.edu/files/files/084nuisancevariables.pdf.

If the way in which your sampling procedure lead to differences in age it presumably may have lead to differences in other things, measured and unmeasured, and including your outcome variables. There has been a lot of work on matching in recent years (https://www.amazon.com/Observational-Studies-Springer-Statistics-Rosenbaum/dp/0387989676), but before looking at these procedures it is necessary for you consider what went wrong with your sampling. This can help you to think about what other biases may exist.

One could use a chisquare test to test whether more (or less) students succeeded who attended the online program than those who attended the traditional program.

The following publications may further help:

I agree with Dennis Njung'e and Aruditya Jasrotia totally. Convenience sampling would do as well.

Hello Matteo,

It wasn't clear why you would be unable to use both an attainment and an importance rating in your model. Maybe you could explain why that restriction exists for your research.

However, it is important to note that difference scores are always less reliable than either of the two constituent scores unless: (a) both measures are perfectly reliable (highly unlikely in your scenario); or (b) the correlation between the measures is zero (probably unlikely as well). If both measures have comparable variance, then the formula for reliability of a difference score is: (mean reliability - rxy) / (1 - rxy), where: mean reliability is the average reliability of the two measures, and rxy is the correlation between measure "x" and measure "y".

One final thought: Imagine two respondents, one with maximum rating on attainment and maximum rating on importance (yielding a zero difference, following your proposed approach); the other with a minimum rating on attainment and minimum rating on importance (again, zero difference). Numerically, these are "identical" cases. However, I'd be hard pressed to argue that they're comparable with respect to the goal in question.

Good luck with your work.

Your request for a convenient research methodology cannot be answered without more information about the context of your research. The choice of a quantitative approach , qualitative approach or a mixed-methods apprach is only the function of what you want to achieve from your research. The gender inequality construct is responsive to multiple theoretical and practical traditions in research. One option to research gender inequality is to opt for quantitative approach that is further developed into a survey design ( Creswell, 2003 ). You need to be careful about the imperatives of a survey design in terms of sampling, data collection and data analysis. You may opt for probability sampling ( your best option if possible) because the representativeness of your population sample is consequential at all levels in survey designs. The instrument for data collection coud be a structured questionaire. As for the data analysis part, you may consider all statistical tools with the capacity to equate correlation, regression and variance.

This one option among many. The important thing is to conduct research from the perspective of an establish tradition in terms of methodology and design.

Best of luck

Two considerations: (1) Many small studies lacking power can be combined to derive inferences with considerable power, so a small study can contribute to knowledge. If there is a deep literature, a Bayesian estimate with an informative prior might be useful. Otherwise, a small study might contribute to meta analysis. (2) If the literature is nonexistent or shallow, think of the study as being exploratory and hypothesis generating, more a qualitative than quantitative analysis, and do not oversell the findings.

done. thank you for such good survey i have start analyzing the mental health of mine due to this corona virus.

Like other members said the MC application is very broad in scope. First, you should consider what kind of project management problem will be solved using MC, after that you can choose the industry. After you decided, you can start to create the system model, describe the scenario (how the system work) and the parameter of input, process, and output in sequence. The important thing about using MC is the model. You can make an assumption for some condition to make a model more logically (similar to the actual condition but not exactly the same, it is very difficult). Make sure you can measure important parameters from each step from your model. Finally, you can simulate the model to get the result. Validate your model with historical data (please don't generate the data by your self, using some reference from database is necessary), make some adjustments to reduce deviation and then you can use it to predict the future situation (at this step you can generate data by your self using what-if analysis based on the industry situation). To simulate the model you can select appropriate software or you can use excel for a simple model. I hope It can help you.

Can you clarify if you are analyzing a scale per se, that is values composed of several Likert items, or if you are analyzing individual Likert items? The recommended analyses will likely be different.

attached file may help....

Basically asking how governmental regulation related to climate change affects economic development?

The big challenge here is how do you know what would have happened if those policies did not take effect. The results could also depend greatly on what part of the economy one looks at. Coal mining and coal use would decline under climate change policies while wind, solar, hydroelectric, and similar sectors will benefit. While individuals will win or lose, the net effect on society is unclear.

Hi Hayley,

You will first need to decide how you will measure 'influence' - and 'outrage'.

You can do this both qualitatively and quantitatively.

If you can specify parameters for influence (e.g. a higher proportion of articles) with outrage (was there a protest at the time? Or some measure of public outrage) - then you could see if the two measure correlate.

A qualitative analysis would definitely be useful.

Again - how you define your constructs will determine how you measure this - and you can then build a rich case study around this :)

I hope this helps.

It's fine to use two distinct models with the same data to test two different research questions. If you want to test the interrelationships between the two dependent variables, then you need to do more, but that depends on how much you expect the two dependent variables to be interrelated. Also, even if you expect the two to be interrelated, testing two independent models might be a good first step (depending on what you wish to test).

Hi Emil,

covariances, regardless whether they are between exogenous variables or between the errors of endogenous variables, reflect non-causal connections between the respective variables. Depending of the exact literature and its history, the meaning of covariances may differ (in SEM, they represent a "don't know-situation", in graph theory/DAGs, they represent omitted confounders) but both interpretations clarify that their connection don't count as "due to Y" (where Y is a respective exposure). If you think that the covariance between X and Y is because Y affects X then substitute the covariance with an effect and incorporate this effect into your indirect effect calculation. Whether this change is legitimate and leads to a correct estimate of the true overall effect of Y, depend of cause whether this change is correct. It is as always: You get estimates that are conditional of the correct specification. Further, you will see that the fit won't change, hence, both models cannot be differentiated with the current data.

The same is true (as Mark noted) when we talk about covariances between two mediators (say, m1 and m2). If you don't think that m1 causes m2, than an omitted/not-estimated error covariance (as Mark long did :) implies that all of the covariance between m1 and m2 will be due to the joint influence of their antecedents. If there are, however, further common causes of both (which are not in the model), you *have to* estimate the error covariance. Otherwise the model won't fit, as the algorithm cannot achieve re-producing the empirical covariance between m1 and m2 (i.e., it is larger than implied by your model). If however, the covariance between m1 and m2 is NOT due to omitted common causes but due to m1 causing m2, than an error covariance would be the wrong thing and the effect of m1 on Z would be underestimated (because its indirect effect via m2 is not present).

Your initial problem of not recognizing that two seemingly equal predictors (X and Y) have indeed a causal structure is called "overcontrol" and refers to controlling for a mediator in order to estimate a variables total effect, see

Elwert, F. (2013). Graphical causal models. In S. L. Morgan (Ed.), Handbook of causal analysis for social research. (pp. 245-273). Dordrecht Heidelberg New York London: Springer.

Shrier, I., & Platt, R. W. (2008). Reducing bias through directed acyclic graphs. British Medical Journal, 8(70), 1-15.

HTH

Holger

If you use statistical models (like Ttest, ANOVA, Pearson r, regression analysis, path analysis, SEM, among others) to test the hypotheses of your study, then I suggest you conduct a statistical power analysis in computing your minimum sample size. Sample size is a function of the following components: effect size, errors in decision (Type I and Type II), complexity of the statistical model, among others. Statistical power analysis or simply power analysis is finding the optimal combination of the said components. You can use the G*Power software which is downloadable for free. Just search it in google.

Slovin's formula has been taught by "irresponsible professors" in the Philippine Colleges and Universities. Sorry for the strong words, but that is true. Because they dont have formal training about Statistics, they taught the wrong things to their students. They are like blind people guiding another blinds.

Anyway, you may read a published article titled "On the misuse of Slovin's Formula".

I am on travel now. I am just using my phone to reply your message. I will email to you later the said article if you want. If interested, just email me at johnny.amora@gmail.com and then I will send you some materials including the said article.

Hi,

to the contrary, mediation is not a statistical concept but a causal concept. A mediator is supposed to transmit the effect of an exposure variable to the outcome. A moderator is a contextual variable on which the occurrence of an causal effect (or statistical relationship, if you broaden the perspective) depends. According to the level of the moderator, the effect can be existent or non-existent or vary with regard to its strength (going from zero to "strong") or vary from negative to positive.

That having said, a mediator can also be a moderator, when the exposure has both a direct and indirect effect (via the mediator) and the mediator moderates the direct effect in its strength.

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51(6), 1173-1182.

Best,

Holger

Thanks Kelvin.

Wishes .

I follow answers

best regards

Biomedical engineering.

Hope this helps,

Matt

I would argue for emphasizing your research questions and the appropriate methods for addressing them. Meanwhile, leave metaphysics to the philosophers.

Thanks Víctor, will try to check it in the future.

Dear Kanika,

Ranked data is not independent. A better test would be Friedman's ANOVA.

Sample size determination is one of the gray areas in research. For qualitative study, you should proceed to include a participant until you reached a saturation point , a point where any more participant hardily generate new information. In your case , you took 4 mangers. If you get sufficient information from them, it could be suffice. For quantitative ones , different methods including various formulas are suggested. The total number of target population, budget, homogeneity and accessibility of your participants should also be considered. To say 40-50 employees is suffice, you need to take these factors into considerations. Please see Krejcie and Morgan sample size determination method (1970) for further information.

If you do not have multiple measurements for any of your variables, then SEM is not appropriate, and should should use regression.

Hi Bruce.

In fact all depends upon the theory you use (actually statistics is always a tool and can be totally meaningless if you don't have a good theory behind the model).

If the theory says a construct A is a 2nd order one, compound by constructs B and C (let's call them "subconstructs" from now on; this is more a poetic license than a real used word) and your results confirm this, that's OK. But, if the model doesn't fit in this approach you should try directly the subconstructs in place of the 2nd order construct.

Attached is a paper I published where the model has some 2nd order constructs but I decided to check if this was the best approach and try some 1st order alternative (actually 3 alternative models). At the end I confirmed that 2nd order model was the best.

In the other way around, if your theory treats the constructs as 1st order (it seems this is your case), there is no reason to try a 2nd order approach, even because what would this 2nd order construct be ?.

If your variables were presented in some papers as a 2nd construct, them you should try; if not, forget about and keep 1st order approach.

See my paper, remembering that, in my case, theory says there are some 2nd order constructs and I decided to check if this was the best approach much more because in Social Sciences sometimes things are not straitghforward.

Hope this can help you.

Any doubt, feel free to call me back.

Regards.

Thanks for providing the link to your paper abstract. If possible, I would like to take a look.

My current plan is to use the CO2e emissions of something as close as possible to the average UK single-family household for comparison. I see this as giving some useful context to the data I will gather from the cohousing communities which I hope to work with. However, I do wonder if I should try and find a more like-for-like comparison by matching certain variables, as you say. For example, cohousing residents tend to come from a certain socio-economic background, and as socio-economic background is strongly correlated with CO2e emissions, perhaps I should be taking this into consideration when deciding upon what secondary data on single-family households to use as comparison.

Hi Jesca,

Thanks for raising the question. I understand that you like to analyse the factors affecting " Food manufacturing SMEs in Tanzania." In this case, I would take both qualitative and quantitative method at different stages of research work. Here is a proposed plan for analysis (Example, testing a proposed model with SEM and any other multivariate technique)

1) Secondary research: Perform a through literature review to identify factors which are affecting " Food manufacturing SMEs" in Tanzania or in other countries.

2) Once you identify the factors, look for the operationalized definition of those constructs. If you get existing scale, you can use those scales with suitable modifications if any for using those in your target country.

3) If you do not find any suitable scale, you can develop a new scale to operationlize the construct. But, It is a lengthy process.

4) Qualitative research (Expert interview): Suppose, you find questions from the previous studies to measure the target factors (constructs), still it is essential to perform content/face validity of those questions in your context seeking expert opinion (you can seek views from few SCM professionals)

5) Quantitative test: Once you fine tune your questionnaire based on the previous feedback (step 4), you can conduct pilot test to check the reliability and validity of measurement instruments. If you find satisfactory results in pilot test, you can move to the final study.

6) Quantitative test: Now, You can conduct quantitative analysis for validate your hypotheses.

Hence, based on your research objective you can use both qualitative and quantitative method.

Hope this helps.

Thank you,

Gobinda

Hi Edita,

Here's a link that should provide some insight: https://books.google.com/books?id=KQ25WExx5usC&pg=PA678&dq=mann+whitney&hl=en&sa=X&ved=0ahUKEwig3vHBlK_aAhUixYMKHY89Bm0QuwUITTAG#v=onepage&q=mann%20whitney&f=false

However, I concur with the preceding responses. Your sample is sufficiently large enough to perform some form of parametric analysis.

Have a great day!

--Adrian

Hi Rokhsareh

Report RMSEA and CFI, that's well enough.

jk

Social security is all-embracing and so is a very wide concept. What does it mean and how is it measured? You may need to narrow its scope to apply it to India. Indian security has its peculiarities? The scope will have to be narrowed down further to apply it to manufacturing industry in India. This will help in the formulation of the research questions and hypotheses.

Very few people use true "convenience" sampling because that implies taking just about anyone. At a minimum, you already have eligibility criteria, and you have set up the possibility of systematic comparisons, so you are doing purposive sampling by definition. What you don't have is random selection within your subgroups of interest, which means your statistical results will not generalize to the population.

You can begin by reading some papers such as :

Sample size calculation is studied for probabilistic sampling. However, if one makes an uncertain hypothesis as that the hazard reproduces probabilistic sampling, then the formulae given for simple random sampling with replacement (for example) could be an approximation to nonprobabilistic quota sampling. This is not scientific completely but it is an approximation whem it is impossible to use probabilistic sampling.

Hi Eltion,

I have written an ado that generates plots for interaction terms of multiplicative regressions. Just type in ssc install interactplot and look up the help file via help interactplot. I provided some simple and some more advanced examples in the help file, which should help you getting used to the command.

Alternatively, you can create marginal effect plots in Stata by using the margins and marginsplot command.

In a strict sense, capital expenditure, capital formation and investment seem to have a little bit difference.

Capital formation is an investment in a newly produced asset. This word may be matched with numbers at the national level.

Capital expenditure is an investment in a newly produced asset as well as exising asset. This word may be more matched with numbers at the sectoral level under the nation. If the analysis is done for households or businesses, capital expenditure will be more relevant.

Investment is a kind of acquisition of assets. The asset can be financial or non-financial. Capital formation and capital expenditure are more relevant to non-financial produced assets.

Therefore depending on the analytical scope, data should be carefully selected, in my opinion.

- The sample size seems adequate for conducting CFA (see e.g. Todd Little's Longitudinal CFA book in which he shows that above N = 120 estimates stabilize in similar models).

- If there are at least 5 categories on each item, use MLR (see http://psych.colorado.edu/~willcutt/pdfs/Rhemtulla_2012.pdf).

- Poor model fit: Look for the reasons (examine residual covariances/correlations, and modification indices) - in case there are obvious reasons, modify model; in case not, leave it as it is. Latent correlations from CFA should be interpreted in any case. Switching to classical correlation analysis wouldn't circumvent the problem of badly fitting model, just aggravate.

The decision whether you use a quantitative technique or qualitative technique is based on the research question; it is not the choice of the researcher. However, if research questions guide you to opt quantitative technique, then there are different methods like an experiment, case study, survey etc. (See the link given below)… Measuring perception is more accurately measured in questionnaire-based survey study. To adopt/adapt/develop a questionnaire, you need to understand the variables of your study, operationalize these and make a decision about selection of questionnaire and scale.

It is not something that I am aware of in the UK. However, there is currently a transition in police officer entry standards in the UK driven by the College of Policing. They may have some background research. However, as policing functions differ hugely I agree with Ben measuring performance will be very difficult. So much so key opinion formers differ enormously on how to do so. Good luck Sean

I agree that the key is the kind of question(s) that you want to answer, but it is also important to consider the kind of skills you want to develop. Often, it is hard to achieve a high level of expertise in both qualitative and quantitative methods, so if you have preference for one or the other, that can be an important part of your decision.

Also, don't overlook the possibility of mixed methods research, where you combine both qualitative and quantitative methods in the same project.

If your Likert scales are ordinal rather than interval, then you need to consider nonparametric analyses. That said, some scales may be interval, some may be ordinal, and while I recommend to be conservative and use non-parametric analyses, other researchers do use parametric analyses for a mixed data-set. However, I think the best practical advice is to find a peer-reviewed journal article that uses your methodology, or similar, in your field and then use this standard as a starting point to organize your analysis work.

If using the SPSS software package I have attached a link to an SPSS manual which is very widely used, and which will contain the decision making steps for statistical test choice and the step-by-step instructions for computing, interpreting, and writing up your analysis section for your results.

Thanks for your reply. I'll try to be more specific.

In my study, I have collected tweets from 4 parties, creating one corpus composed by 4 sub-corpora. Then, I used a software to annotate tweets following the Appraisal framework which consists of three main features: graduation, attitude and engagement. The four examined parties are expected to exploit graduation of meaning and attitude expression more frequently than other parties (I am also using reference corpora), while at the same time showing less propensity to engage with different points of view. Hence, the hypothesis are the following:

H0: There is no significant difference in graduation, attitude and engagement instances between populist and democratic parties tweets.

H1: Populist parties tweets present a higher number of graduation and attitude occurrences, together with a lower frequency of engagement instances, when compared to reference corpora.

The level of significance for the hypothesis was tested at a 1% level of significance (i.e., α = 0,01). In order to reject the null hypothesis, the relative frequency values regarding graduation and attitude have to be to be ≥1.20, while the same value for engagement is supposed to be <0.80. These numbers are relative frequencies automatically calculated by the software, which compare occurences of the three features in both corpora of interest and reference corpora.

The following papers should be helpful: 

Please let me know if this reference (freely downloadable) is helpful to you:

Download site:  http://www.jwalkonline.org/docs/Grad%20Classes/Survey/articles/required/Podsakoff%20-%20common%20method%20biases.pdf

Philip M. Podsakoff, Scott B. MacKenzie, and

Jeong-Yeon Lee

Indiana University

Nathan P. Podsakoff

University of Florida

Interest in the problem of method biases has a long history in the behavioral sciences. Despite this, a comprehensive summary of the potential sources of method biases and how to control for them does not exist. Therefore, the purpose of this article is to examine the extent to which method biases influence behavioral research results, identify potential sources of method biases, discuss the cognitive processes

through which method biases influence responses to measures, evaluate the many different procedural and statistical techniques that can be used to control method biases, and provide recommendations for ...

Dennis

Dennis Mazur

Please let me know if these references/sites are useful to you:

1.  [PDF]Maximum likelihood estimation of endogenous switching regression ...

by M Lokshin - ‎Cited by 287 - ‎Related articles

In this model, a switching equation sorts individuals over two different states (with one regime observed). The econometric problem of fitting a model with endogenous switching arises in a variety of settings in labor economics, the modeling of housing demand, and the modeling of markets in disequilibrium.

2.  [PDF]SWITCHING REGRESSION MODELS AND ESTIMATION

Outline. Switching Regression Models. Model setting. Motivation. Estimation ( Two-stage method). Variations. Censored models. Models with self-selectivity ...

3.  Endogenous Switching Regression Models with Limited Dependent ...

by DA POWERS - ‎1993 - ‎Cited by 10 - ‎Related articles

Social research often involves estimating the effects of a categorical treatment on a dependent outcome variable. Endogenous switching regression models are natural extensions of classical experimental designs, which allow tests of assumptions about the exogeneity of treatment effects from survey data.

Dennis

Dennis Mazur

Good question. I think this shows one of the limits of face validity. Who are the experts?

If it does not seem "valid" for different groups of users (is users a better group than experts?), that does not mean it is not valid in other senses of the word validity (e.g., causal validity by Borsboom, Kane's how it is used validity, Meehl and Cronbach, etc.). One problem with experts is that they are not always right, This is why the US courts no longer use just general acceptance (Frye) for accepting expert testimony (the Daubert trilogy of cases addressed this). Galileo was not helped by "experts."

It sounds like you have a mixed methods design of the from QUAL --> quant. Depending on your purposes, this can be quite useful, as I have argued in the attached publication.

Still, I have to agree with Daniel Wright that the term "confirm" can be problematic, especially when you are dealing with the results from qualitative research. So, you should be sensitive to the preferences of your audience when you choose the language you use.

Hi Emma,

"Common variance, or the variance accounted for by the factor, which is estimated on the basis of variance shared with other indicators in the analysis; and (2) unique variance, which is a combination of reliable variance that is specifc to the indicator (i.e., systematic factors that influence only one indicator) and random error variance (i.e., measurement error or unreliability in the indicator)." (Brown, 2015).

it can be said that If the factor loading is 0.75, observed variable explains the latent variable variance of (0.75^2=0,56) %56. It is good measure. So if your factor loading is 0.40, it explains %16 variance. As a cut point 0.33 factor loading can be given. Beacuse of it explains %10 variance.

Not a survey. We are testing whether CG codes issued in 2002 by capital markets authority (CMA) was able to deter earnings management (EM) in Kenya for non-finance listed firms. We develop a CG index based on the provisions of the CG code and apply Jones model to determine discretion accruals regarded as a proxy for EM. Our contribution is based on applying composite CG index contrary to existing studies. Our study also provides support to the initiatives by CMA to enact a new code and other changes at the NSE. Our interest is to assess the effectiveness of CG code issued in 2002 under conditions described as “developing country situation”. We could not find CG data for firms in other EA states but our findings can be generalized to the common law developing countries.

The project is closed. Look out for a special issue on Africa by the of the Journal of Accounting in Emerging Economies that will be out any time now.

General rule: EFA > Used for instruments (or scales) that have never been tested before (for their validity are reliability). CFA > Used for instruments (or scales) that have been tested before (for their validity are reliability).

I argue that when you translate an existing instrument (or scale) (that has been tested before for its validity are reliability), in order to use it in another country (different language), this instrument (scale) becomes “new”. So, you need to perform EFA.

Moreover, I argue that when you use an existing instrument (or scale) (that has been tested before for its validity are reliability) in the same country (same language), but in another sector or research setting, this instrument (scale) also becomes “new”, since it is being tested (used) on a very different population / sample. So, you also need to perform EFA.

So, when should we use CFA? My point of view is that CFA should be used in empirical studies that use instruments (or scales) that have been tested in many previous studies (instruments or scales that have extensively being tested by their validity and reliability). Then, our only job is to confirm that these instruments (or scales) are valid and reliable in our research setting.

On a personal note, I tend to perform both analyses: first EFA, then CFA. I consider that the validity and reliability of my instrument is enhanced with this dual approach.

Thank you very much for the help. Unfortunately, when authors describe descriptive data by median and IQR it is usually because their scores are not normally distributed.

Hi Saraswati

Peter's advice is excellent. If you decided to power up for a medium effect, then you would need 64 patients in each study arm for a 2-arm trial with 80% power and alpha=0.05. If as Peter suggested, you go for a 3-arm trial, then you would need 69 patients per study arm to allow for (say) two pairwise comparisons and with alpha set to 0.04 to adjust for multiple comparisons. Although choosing a medium effect size might seem a bit arbitrary, a difference of 10 on a 100-point scale (eg the SF-36) has often been found to be of clinical importance. Further, a 100-point scale is likely to have an SD of around 20. This gives an effect size of 0.5 (or medium), which is what we have powered for.

Adrian

 Hi Mazhar Ali,

Always the statiscal proceedings are polemic! And in the science we have many opinios.

There is justified if your scale is "virgin" and yet not supported by theory. (generated by Interviews or only researchers´ opinions) The name exploratory means exploration of data, when us do not have a theoretical factorial model. But we never applied EFA and CFA in same sample!

We usually split the same sample into two random groups. With the first, we do the AFE and with second sample we do the AFC.

EFA "...eventually leads to factors which the investigator then interprets. It tends to be stepwise (datadriven) rather than direct (theory-driven)" (Nunnally & Bernstein, 1994, p. 450), i. e., CFA.

See:

Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory (3rd ed.). New York: McGraw-Hill. [chapter 11: Exploratory and Confirmatory Analysis p, 450 - ] for a very important explanation.

hope this helps.

Dirceu

Mr. Marco V. Rossi,

Thank you very much for presenting your take.

"Brand experience 1 is an antecedent of the customer engagement state 1 which leads in tun to the brand experience 2 (in this case, a consequence of the state of engagement). It's a circular process according to me."

Ya! It is a pretty convincing comment. 

Cheers!

Which code you will use?

Thank you.

Thank you Kelvyn, this looks very relevant.

Dear Sawitree, everything cs/mv/fsQCA need is a table of suitable data. How these data were produced in the first place is, strictly speaking, not part of QCA itself (this process is called "calibration"; for instance, the UN's Human Development Index uses calibration, but not QCA).

If I understand your question correctly, you'd like to search for particular collocations of words in your interview transcripts and assign corresponding values (e.g., find "hate*", assign "0"; find "love*", assign "1").

What I doubt is that fsQCA is the most appropriate solution for your purposes. By using fsQCA, you essentially lose all categories in your analysis between the two most extreme ones (e.g., everything between "hate*" and "love*"). Given the qualitative nuances of interview data, mvQCA seems to be the much better choice in your case. In mvQCA, all intermediate categories (e.g., "dislike", "indifference", "affection") are preserved in the analysis.

You can easily run mvQCA in my QCApro package (http://www.alrik-thiem.net/software/). Why you should not use the Tosmana software (the software most often used for mvQCA in empirical research) is explained here: https://youtu.be/n8k4OQY5mHg.

Best wishes,

Alrik

There is nothing in the literature to specifiy exactly how many experst as a mimimum or maximum. In my opinion, the most important issue is to identify who is expert in the field. From observing similar issues in the publicaitons, 3-5 experts could be suitable.

good luck

You can compare the means of the continious scale scores of those responding "yes" with those responding as "no" for each dichotomous question. Depending on whether you have normal distribution you can do this by a t-test or a Mann Whitney U test, former being the parametric one. Imho, this seems to be the most appealing way to do it.

Dear Hassan,

I really thank you so much for your generous support,

Cheers

There is a difference between single, Likert-scored questions, which should be treated as ordinal, versus multi-item scales composed of Likert-scored variables, which can be treated as continuous variables.

Because there have been so many questions here about this topic, I have put together a set of resources:

I don't understand why you want to create cut-offs for a continuous variable, which would involve a loss of information. If this variable is typically analyzed using techniques such as correlations and regression, then it doesn't make sense to chop it up.

Dear Jeroen Nawijn, Can I get the fulltext?

Thank you Saurav.

Dear Peter, can I get the link or fulltext?

Hi, I thin because of the Bragg peak behavior and the overall very short reach of Alpha particles it will be quit difficult to get the exact energy value. However, you van try to use a certain distance in air to slow the alpha particles down. Here you can find the 'Bragg curve of 5.49 MeV alpha particles in air' https://en.wikipedia.org/wiki/File:Bragg_Curve_for_Alphas_in_Air.png

and the corresponding formulas in this article:

I hope thats helpful.

Best regards

Tobias

I would suggest using a categorical variable, such as "optimal" (PSQI < 6), "borderline" (PSQI 6-7), and "poor" (PSQI >7). See Lund et al. (2010), which uses these categories. If you use the global score as a continuous variable, you will get too large of a range to analyze the data. For example, poor sleepers can have a PSQI of 19, which would be considered an outlier using a continuous variable. You could always just use "poor" or "optimal/normal" as 2 categories, if you are sure to explicitly operationalize your variables.

Dear Palash,

I appreciate your interest in our work and thank you for your comments.

ROC curve, is a graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied.

The diagnostic performance of a test is the accuracy of a test to discriminate diseased cases from normal controls.

ROC curves can also be used to compare the diagnostic performance of two or more laboratory tests.

ROC Curves plot the true positive rate (sensitivity) against the false positive rate (1-specificity) for the different possible cutpoints of a diagnostic test. Each point on the ROC curve represents a sensitivity/specificity pair.

The closer the curve follows the left side border and the top border, the more accurate the test.

The closer the curve is to the 45-degree diagonal, the less accurate the test.

See also https://www.mailman.columbia.edu/research/population-health-methods/evaluating-risk-prediction-roc-curves

and http://www.scielo.br/scielo.php?pid=S0021-75572009000100008&script=sci_arttext&tlng=en

Hope this helps.

No. it is applicable only for randomized clinical trials in 2 different groups

Hi Shahril,

I have the questionnaire. Feel free to mail me:

Regards,

Josh.

Do you only have smoking as a binary variable, or do you also have amount smoking? If so a zero-inflated model might be worth exploring, but as I said above more information is necessary before suggesting a specific test.

Scoring the age groups at the mid-point of each interval will give you the kind of continuous variable that you want, as will scoring education in terms of years of schooling. In the U.S., we have scores of "occupational prestige" that would cover professions, but I have no idea how well those would work in a different cultural context..

Hi Mark,

Traditionally the use of questionnaire data has been popular in measuring bias and or stereotypes but and there are many well-documented questionnaires in an extensive array of  human bias areas that are readily accessible, however, due to the limitations of self-report data (e.g., unreliability of introspective data, social desirability confounds) a range of implicit measures have been developed in which participants behaviour is deemed to indicate bias. Implicit bias tests are usually time-based and, briefly, if participants respond faster to affirm, for example, thin-intelligent versus fat-intelligent this is deemed a demonstration of pro-thin bias. Bias or stereotype is deemed "implicit" in the participant's behavioural response (measured via response latencies compared for affirmation of biased versus nonbiased responding) rather than participants' report of their own bias or lack of same. The most well-known of these implicit measures is the Implicit Association Test/ (IAT; check out Project Implicit, Brian Nosek's webpage in particular is v. informative). The Implicit Relational Asessement Procedure/IRAP is a similar computerised test programme that was developed by Dermot Barnes-Holmes as a behavioural (rather than associationist) measure of implicit bias; this has been used in a variety of areas, i have used it myself to research areas such as pro-thin bias, pro-beauty bias, gender bias, ageist bias et cetera. Your can make further enquiries about the IRAP by contacting Prof. Barnes-Holmes at Ghent University. I hope this information has been helpful.

Best

Carol